Tony_tech/athena-oracle
3
0
Fork 0
Code Issues 6 Pull Requests 2 Actions Packages Projects Releases Wiki Activity

pre-fix snapshot: oracle pipeline (2026-07-16)

Browse Source
This commit is contained in:
Epictetus
2026-07-16 04:27:29 +00:00
parent 5224645703
commit af11b0952d
43 changed files with 7141 additions and 1558 deletions
Download Patch File Download Diff File Expand all files Collapse all files
Issue_003_aiND_edition.md
+69
View File
@@ -0,0 +1,69 @@
# aiND Edition — 2026-07-15 (Curated)
**Status:** HUMAN-APPROVED (8 stories) · staged in `entries` (curated_by=leonard_aiND, manual_section, manual_tier set)
**Not yet on the live webroot** — see deployment note at bottom.
**Lens applied:** build / numbers / outcome / lesson, else scroll.
---
## What People Shipped
*Community builds, ships, and earnings — the T1 gold.*
**Open-Source Local LLM Training Tool (for consumer hardware)** `(id 2315, T1)`
Train/fine-tune local models on consumer GPUs. Install and run today.
→ *Do: install it. Run on your hardware.*
**I Used AI To Sell 10 Websites This Week** `(id 2648, T1)`
A developer shipped 10 client sites with AI help. Concrete paid outcome.
→ *Learn: what actually closed deals.*
**Show HN: I RL-trained an agent that trains models with RL (~$1.3k)** `(id 2461, T1)`
Full build under a hard cost ceiling. Money spent, artifact shipped.
→ *Do: replicate the $1.3k training loop.*
---
## Benchmarks & Builds
*Real numbers, real methods.*
**I benchmarked 15 "E-Waste" GPUs with Modern Workloads** `(id 1969, T4)`
Cheap-hardware numbers a home-lab operator can compare against.
→ *Do: benchmark your junk-drawer GPUs.*
**GPUHedge: serverless GPU hedging drops cold-start p95 117s → 30s** `(id 2438, T4)`
Measured latency win with a concrete method.
→ *Do: steal the hedging pattern for your GPU jobs.*
**New LLM Coordination Benchmark — Multi-Agent Coordination** `(id 2441, T4)`
A real benchmark with a method.
→ *Do: run it against your own multi-agent setup.*
---
## Problem Solved
*Transferable lessons from the field.*
**Structured output reliability with LLMs — 3-month production learnings** `(id 2656, T5)`
What held up, what broke, over 90 days of prod.
→ *Do: apply the reliability pattern to your agents.*
**The absolute nightmare of putting AI agents into actual production** `(id 2645, T5)`
What breaks when agents hit prod. A lesson, not a feature.
→ *Do: pre-empt the failure modes.*
---
## Worth Trying Tonight (highlight)
1. **Install the Open-Source Local LLM Training Tool** `(id 2315)` — highest-signal build this window.
2. **Replicate the $1.3k RL-trained agent** `(id 2461)` — capped cost, full artifact.
---
## Editorial notes
- 8 of 18 reviewed → ~44% of this curated window. The 10 rejects were: PalmClaw (2627), Jacquard (2222), BillAI Bass (2463), NN-in-SQL (2226), Agent-identity (2314), CoT scaling-trap (2444), Upgrade-path (1982), Ford-350 (2649), Codex-encrypt (2157), Ghostcommit (2327).
- Calibration locked: the editor (you) is stricter than the algorithm. Real build / real measurement / real outcome / real lesson — or it scrolls.
- Age: stories ingested Jul 13–14. Not "today," but none were previously posted (recency guard clear). Suitable for a 48h edition.
## Deployment note (BLOCKER — not silently worked around)
The live webroot is served by `render_site.py` / `propagate_stack_now.py`, which output the **legacy "Athena AI News — Clickability"** product. They rank by virality and **ignore** `manual_section`/`manual_tier`, so pushing now would either drop these 8 or bury them in a feed that contradicts the aiND brand. Three webroots exist (preprod1/2/3), target unconfirmed.
**Next step requires a decision:** build a minimal aiND renderer that reads `manual_section`/`manual_tier` and emits the sections above (separate from the clickability product), OR confirm the intended webroot. Nothing was pushed to any webroot.
adapters/_store.py
+2 -2
View File
@@ -91,7 +91,7 @@ INSERT INTO entries
(source, source_id, url, title, extracted_text, summary,
category_tags, signal_score, raw_metadata, first_seen, last_updated)
VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
ON CONFLICT(url) DO UPDATE SET
ON CONFLICT(source, source_id) DO UPDATE SET
source = excluded.source,
source_id = excluded.source_id,
title = excluded.title,
@@ -101,7 +101,7 @@ ON CONFLICT(url) DO UPDATE SET
signal_score = excluded.signal_score,
raw_metadata = excluded.raw_metadata,
last_updated = excluded.last_updated,
first_seen = COALESCE((SELECT first_seen FROM entries WHERE url = excluded.url), excluded.first_seen)
first_seen = COALESCE((SELECT first_seen FROM entries WHERE source = excluded.source AND source_id = excluded.source_id), excluded.first_seen)
"""
aind_edition_2026-07-15.md
+99
View File
@@ -0,0 +1,99 @@
# aiND Edition — 2026-07-15 (Annotation System v1)
**Status:** ✅ APPROVED — 15 stories (user authorization, approval gate cleared)
**Treatment:** aiND Story Annotation System applied (Class A / Class B + visual markers)
**Lens:** build / numbers / outcome / lesson, else scroll
**Window:** released 2026-07-14 / 07-15 (age-dominant; recency-guard clean)
**Not yet on the live webroot** — the legacy generator (preprod1/2) renders virality-ranked CFP-clone output and cannot display annotations. An aiND renderer (Path B) is required to publish this edition visually. See bottom.
---
## 💻 Local AI — Open Source Local LLM Training Tool (consumer hardware)
**Summary:** A new open-source tool lets you fine-tune and train LLMs on consumer-grade GPUs — no datacenter required.
**Why It Matters:** Local training has been locked behind cloud APIs or expensive hardware. This closes the gap for solo builders.
**Builder Takeaway:** Clone it, point it at your dataset, and run a LoRA on the GPU you already own.
## 🏆 Builder Outcome — Show HN: I RL-trained an agent that trains models (~$1.3k)
**Summary:** A builder RL-trained an agent that trains other models, end to end, for about $1.3k in compute.
**Why It Matters:** It shows the full RL-training loop is reachable on a personal budget — not just frontier labs.
**Builder Takeaway:** The $1.3k number is the blueprint. Steal the approach before you scale spend.
## 🏆 Builder Outcome — I Used AI To Sell 10 Websites This Week
**Summary:** A developer used AI to ship and close 10 client websites in a week.
**Why It Matters:** This is the Builder Outcome the whole publication exists for — real money, real delivery.
**Builder Takeaway:** The leverage wasn't the code. It was using AI to run the full sales-to-delivery loop.
## ✅ Production Lesson — Structured output reliability (3-month production learnings)
**Summary:** Three months of hard-won lessons on getting LLMs to emit reliable structured output in production.
**Why It Matters:** Structured output is the unglamorous backbone of every agent that does real work.
**Builder Takeaway:** The failure modes are predictable. Read this before you wire your next agent to a schema.
## 🛠 Builder Tool — deja-vu: open-source memory for coding agents (SSH sync)
**Summary:** A self-hosted memory layer for coding agents that syncs over SSH — your agent remembers across machines.
**Why It Matters:** Agent memory is usually locked in a vendor's cloud. This keeps it local and portable.
**Builder Takeaway:** Drop it into your coding-agent setup if you want persistent context without a SaaS dependency.
## 👀 Worth Watching — Inkling: Open-Weights 975B LLM
**Summary:** Thinking Machines released Inkling, a 975B-parameter open-weights model.
**Why It Matters:** Open-weights at this scale is rare — but 975B won't run on your laptop, so the "why care" isn't obvious.
**Builder Takeaway:** Watch the licensing and quantization path. If it gets distilled or quantized, it becomes relevant to self-hosters.
**Editor's Note:** We're covering this because open-weights at frontier scale is a signal worth tracking, not because you can run it today. Most builders won't touch 975B locally yet — but the weight release itself changes what's possible downstream. Pay attention to who forks it and what small versions appear.
## 🛠 Builder Tool — Doodles → charcoal vector agent
**Summary:** A builder made an agent that turns Remarkable doodles into clean, editable charcoal-style vector art.
**Why It Matters:** It's a real, shippable creative tool — not a demo. The output is actual vectors you can edit.
**Builder Takeaway:** The pattern (sketch → editable asset) applies to any design workflow. Fork the idea for your own medium.
## 🛠 Builder Tool — [P] RL-training Qwen3.6 to RL-train tool
**Summary:** A builder posted a reproducible setup for using Qwen3.6 to RL-train a tool-using model.
**Why It Matters:** Reproducible RL-training recipes are scarce; this is one you can actually run.
**Builder Takeaway:** If you've wanted to train a tool-using agent, this is a starting point with real numbers.
## 📊 Benchmark — GPUHedge: cold-start p95 117s → 30s
**Summary:** Hedging across serverless GPU providers cut cold-start p95 from 117s to 30s.
**Why It Matters:** Cold starts are the silent tax on every GPU-backed service. This is a measured fix.
**Builder Takeaway:** If you serve inference serverlessly, the hedging pattern is copy-paste worthy.
## 📊 Benchmark — LLM Coordination Benchmark (multi-agent)
**Summary:** A new benchmark for measuring how well multiple language agents coordinate on open-ended tasks.
**Why It Matters:** Multi-agent systems are hot but unmeasured. A benchmark is the first step to knowing if your setup actually coordinates or just chatters.
**Builder Takeaway:** Run it against your own multi-agent stack to get a baseline before you optimize.
**Editor's Note:** Benchmarks are easy to over-hype and easy to ignore. This one matters because coordination — not capability — is the bottleneck most multi-agent builds hit first. You won't implement it tomorrow, but you should know your agents' coordination score before claiming they "work together."
## 📊 Benchmark — ViHoRec: Vietnamese hotel rec dataset + cold-start benchmark
**Summary:** A quality-controlled Vietnamese hotel recommendation dataset with a cold-start benchmark.
**Why It Matters:** It's narrow (hotel rec, Vietnamese), but it's a clean, usable dataset — useful if you work in recsys or low-resource languages.
**Builder Takeaway:** If your domain is recommendations or non-English, this is a ready training/eval set.
**Editor's Note:** We included this because it's a rare thing: a dataset that's actually clean and immediately usable, not another paper promising one. Most builders won't need Vietnamese hotel recs — but the methodology for building a quality-controlled benchmark is transferable to your own niche.
## ✅ Production Lesson — Form, Not Content? (self-repair eval)
**Summary:** A placebo-controlled study asking whether "self-repair" in small code models comes from the form of the prompt or actual learned behavior.
**Why It Matters:** Self-repair is sold as a model capability. This study suggests the gain may be shallower than it looks.
**Builder Takeaway:** Don't assume prompt-based self-repair generalizes. Test it on your own failures.
**Editor's Note:** This looks like research on the surface, but the real story is the production lesson hidden inside it. Pay attention to the methodology rather than the paper itself — it tells you how to tell real self-repair from prompt theater.
## ✅ Production Lesson — The Illusion of Robustness
**Summary:** A study showing aggregate accuracy hides prediction flips under task-irrelevant context.
**Why It Matters:** Your model can look 95% accurate and still flip its answer when you add irrelevant text.
**Builder Takeaway:** Evaluate on perturbed inputs, not clean benchmarks, before you trust a model in prod.
**Editor's Note:** The interesting part isn't the benchmark. It's what the benchmark reveals: aggregate scores lie about edge behavior. If you ship a model on a single accuracy number, this is your warning shot.
## 🛠 Builder Tool — quantumbyte: open-source app builder engine
**Summary:** An open-source engine that turns intent into working app workflows.
**Why It Matters:** Another builder-focused tool in a space full of closed SaaS. Open and extensible.
**Builder Takeaway:** If you're building internal tools or agent workflows, this is worth a look as a foundation.
---
## 🛠 Builder Tool — pxpipe: cut Claude token cost ~70% by sending context as PNGs
**Summary:** A local proxy (127.0.0.1:47821) renders bulky system prompts, tool docs, and old history into PNGs so they bill by pixels (~3.1 chars/token) instead of text (~1 char/token), keeping recent turns as plain text to protect quality. Open source; 6k+ stars and ~70% savings are the project's/X-post claims, not yet independently verified.
**Why It Matters:** System prompts, RAG context, and conversation history are the biggest token cost line for most Claude builders. This attacks exactly that — without changing your app code.
**Builder Takeaway:** Run it locally as a drop-in proxy, let it image-ify the stale context, keep recent turns as text, and watch the live estimator. Validate the ~70% on your own workload before trusting it — vision tokenization is resolution/tile-based, so the real ratio depends on your content.
## Editorial treatment summary
- **Class A (standalone, no note):** 11 stories — 2315, 2461, 2648, 2656, 3166, 2326, 2429, 2438, 2735, pxpipe (3208), and the three Production-Lesson/Benchmark A's above (2637, 2638, 2656).
- **Class B (Editor's Note added):** 4 stories — 3167 (Inkling), 2441 (Coordination Benchmark), 2640 (ViHoRec), 2638 (Form Not Content). Note: 2637 (Illusion of Robustness) and 2656 (Structured output) are Production-Lesson Class A.
- **Visual markers used:** 💻 Local AI · 🏆 Builder Outcome · ✅ Production Lesson · 🛠 Builder Tool · 👀 Worth Watching · 📊 Benchmark.
## Deployment note (BLOCKER, not silently worked around)
The live generator (`generate_from_athena.py`, preprod1/2) ranks by **virality** and emits the **"AI NEWS DAILY" CFP-clone** layout. It has **no field for annotations, class, or markers** — it cannot render this edition. To publish visually, build the **aiND renderer (Path B)**: read `aind_class` / `aind_marker` / annotation columns, emit sections with markers + Editor's Notes. Until then, this file is the publication-ready artifact and the 14 are approved in `entries` (curated_by=leonard_aiND, aind_class, aind_marker set).
athena_llm_review_prompt.txt
+217
View File
@@ -0,0 +1,217 @@
You are the founding editor of 'AI Enthusiast Daily' — a publication for AI builders / practitioners, NOT the general public.
Below is a Sprint 1 rules-based classification dump of 200 stories (id, source, bucket, final_score, title, url).
For EACH row, decide ONE of: PUBLISH / REJECT / BORDERLINE — from the single question:
"Would an AI enthusiast proudly read this and learn something actionable?"
Heuristics:
- REJECT by default: lawsuits, IPO/valuation gossip, CEO opinions, celebrity AI takes, vague culture pieces, pure funding rounds with no technical substance.
- PUBLISH leans toward: local-model how-tos, benchmarks with real numbers, reproducible tooling, agent infra with code, novel research with a clear builder angle.
- BORDERLINE: technically adjacent but thin, or depends on execution quality.
- Do NOT change the bucket. Do NOT trust final_score — it is a draft signal only.
- This is a PROPOSAL for human confirmation; the human editor makes the final call.
Output ONLY a table, no commentary:
id | verdict | one-line reason
=== DATA (tab-separated: id source bucket final_score title url) ===
2702 rss SHIPPING 0.180 OpenAI researcher Miles Wang in talks to launch AI drug discovery startup valued at $2B https://techcrunch.com/2026/07/14/openai-researcher-miles-wang-in-talks-to-launch-ai-drug-discovery-startup-valued-at-2b/
2587 rss CULTURE 0.000 Lorde says AI glasses are ‘not sexy’ https://techcrunch.com/2026/07/14/lorde-says-ai-glasses-are-not-sexy/
2584 rss UNCATEGORIZED 0.030 OpenAI’s first hardware device is reportedly a screenless speaker that can move https://techcrunch.com/2026/07/14/openais-first-hardware-device-is-reportedly-a-screenless-speaker-that-can-move/
2589 rss CULTURE 0.000 OpenAI pushes back on Apple trade secret lawsuit https://techcrunch.com/2026/07/14/openai-pushes-back-on-apple-trade-secret-lawsuit/
2672 hackernews UNCATEGORIZED 0.020 Financing the AI boom: from cash flows to debt [pdf] https://www.bis.org/publ/bisbull120.pdf
2585 rss MODEL RELEASE 0.120 OpenAI’s new flagship model deletes files on its own, people keep warning https://techcrunch.com/2026/07/14/openais-new-flagship-model-deletes-files-on-its-own-people-keep-warning/
2646 reddit MODEL RELEASE 0.130 Opening the Black Box: Unison Zero Parameter Model https://www.reddit.com/r/artificial/comments/1uwjwl6/opening_the_black_box_unison_zero_parameter_model/
2648 reddit PROBLEM SOLVED 0.220 Developers Hate AI. I Used It To Sell 10 Websites This Week. https://www.reddit.com/r/artificial/comments/1uwj75g/developers_hate_ai_i_used_it_to_sell_10_websites/
2485 rss SHIPPING 0.210 Apple opens its new Siri AI to everyone with the iOS 27 public beta https://techcrunch.com/2026/07/14/apple-opens-its-new-siri-ai-to-everyone-with-the-ios-27-public-beta/
2486 rss CULTURE 0.000 Anthropic’s newest ad is creeping people out https://techcrunch.com/2026/07/14/anthropics-newest-ad-is-creeping-people-out/
2487 rss BUSINESS 0.030 The founder of Hinge raised $18M to build a new AI dating service, Overtone https://techcrunch.com/2026/07/14/the-founder-of-hinge-raised-18m-to-build-a-new-ai-dating-service-overtone/
2654 reddit BUSINESS 0.010 How does a 102M-parameter transformer forecast multivariate time series? https://www.reddit.com/r/artificial/comments/1uwh9ko/how_does_a_102mparameter_transformer_forecast/
2490 rss MODEL RELEASE 0.090 Google faces another AI training lawsuit from major publishers https://techcrunch.com/2026/07/14/google-faces-another-ai-training-lawsuit-from-major-publishers/
2652 reddit CULTURE 0.000 Apple just sued OpenAI for trade secret theft. And Google quietly rewrote how the internet works. https://www.reddit.com/r/artificial/comments/1uwh06x/apple_just_sued_openai_for_trade_secret_theft_and/
2645 reddit SHIPPING 0.230 The absolute nightmare of putting AI agents into actual production https://www.reddit.com/r/artificial/comments/1uwg8kk/the_absolute_nightmare_of_putting_ai_agents_into/
2624 arxiv RESEARCH 0.190 Do AI Agents Know When a Task Is Simple? Toward Complexity-Aware Reasoning and Execution https://arxiv.org/abs/2607.13034v1
2625 arxiv RESEARCH 0.150 The Seriality Gap in Video Diffusion Models https://arxiv.org/abs/2607.13031v1
2626 arxiv RESEARCH 0.340 TerraZero: Procedural Driving Simulation for Zero-Demonstration Self-Play at Scale https://arxiv.org/abs/2607.13028v1
2627 arxiv RESEARCH 0.340 PalmClaw: A Native On-Device Agent Framework for Mobile Phones https://arxiv.org/abs/2607.13027v1
2628 arxiv RESEARCH 0.150 A Shortcut to Statistically Steady-State Turbulence with Flow Matching https://arxiv.org/abs/2607.13022v1
2649 reddit PROBLEM SOLVED 0.210 Ford replaced engineers with AI, then quietly hired 350 back. The reason should stop every founder about to cut their team to SAVE money. https://www.reddit.com/r/artificial/comments/1uwg31g/ford_replaced_engineers_with_ai_then_quietly/
2629 arxiv RESEARCH 0.110 Audio-Native Speech Recognition with a Frozen Discrete-Diffusion Language Model https://arxiv.org/abs/2607.13013v1
2630 arxiv RESEARCH 0.140 Dynamic Resource Allocation for Ensemble Determinization MCTS https://arxiv.org/abs/2607.13007v1
2631 arxiv RESEARCH 0.180 The Spectrum Is Not Enough: When Context Helps Time-Series Forecasting https://arxiv.org/abs/2607.13006v1
2632 arxiv RESEARCH 0.150 Watermark Forensics for Generative Models: An Information-Theoretic Perspective https://arxiv.org/abs/2607.13003v1
2494 rss MODEL RELEASE 0.090 DeepMind CEO calls for an independent standards body to regulate frontier AI https://techcrunch.com/2026/07/14/deepmind-ceo-calls-for-an-independent-standards-body-to-regulate-frontier-ai/
2429 reddit MODEL RELEASE 0.210 [P] RL-training Qwen3.6 to RL-train tool using AI models [P] https://www.reddit.com/r/MachineLearning/comments/1uwfmfa/p_rltraining_qwen36_to_rltrain_tool_using_ai/
2633 arxiv RESEARCH 0.110 Win by Silence: Deletion Non-Monotonicity, Autonomous Exploitation, and Typed-State Gating in LLM Plan Evaluation https://arxiv.org/abs/2607.12986v1
2634 arxiv RESEARCH 0.110 Resist and Update: Counterfactual Report Coordinates for Incentive-Compatible LLMs https://arxiv.org/abs/2607.12985v1
2635 arxiv RESEARCH 0.140 FormalAnalyticGeo: A Neural-Symbolic Based Framework for Multimodal Analytic Geometry Problem Generation https://arxiv.org/abs/2607.12982v1
2488 rss MODEL RELEASE 0.090 Anthropic opens Claude for Teachers with a promise not to train models on student data https://the-decoder.com/anthropic-opens-claude-for-teachers-with-a-promise-not-to-train-models-on-student-data/
2636 arxiv RESEARCH 0.180 Ensemble Controlled-Flow Filtering for Implicit Data Assimilation https://arxiv.org/abs/2607.12975v1
2637 arxiv RESEARCH 0.210 The Illusion of Robustness: Aggregate Accuracy Hides Prediction Flips under Task-Irrelevant Context https://arxiv.org/abs/2607.12963v1
2638 arxiv RESEARCH 0.440 Form, Not Content? A Preregistered, Placebo-Controlled Evaluation of Learned Error-Conditioned Self-Repair Through Prompts and Weights in Frozen Small Code Models https://arxiv.org/abs/2607.12962v1
2656 reddit SHIPPING 0.190 Structured output reliability with LLMs — 3-month production learnings https://www.reddit.com/r/artificial/comments/1uwe9qp/structured_output_reliability_with_llms_3month/
2639 arxiv RESEARCH 0.190 Robustness of Deep Learning Models for PV Power Forecasting under NWP Forecast Errors: A Spatiotemporal and Physically Interpretable Analysis https://arxiv.org/abs/2607.12954v1
2640 arxiv RESEARCH 0.250 ViHoRec: A Quality-Controlled Vietnamese Hotel Recommendation Dataset and Cold-Start Benchmark https://arxiv.org/abs/2607.12946v1
2502 rss BUSINESS 0.030 DeepSeek needs more cash just weeks after closing its first $7 billion round https://the-decoder.com/deepseek-needs-more-cash-just-weeks-after-closing-its-first-7-billion-round/
2168 rss CULTURE 0.000 Meta’s Adam Mosseri says AI token budgets could soon be capped per engineer https://techcrunch.com/2026/07/14/metas-adam-mosseri-says-ai-token-budgets-could-soon-be-capped-per-engineer/
2164 rss UNCATEGORIZED 0.030 Google Search now generates AI images when it can't find what you're looking for on the web https://the-decoder.com/google-search-now-generates-ai-images-when-it-cant-find-what-youre-looking-for-on-the-web/
2150 reddit MODEL RELEASE 0.140 All cross thread implementation of memory in chatgpt, claude, and gemini is unsafe https://www.reddit.com/r/artificial/comments/1uwdc0k/all_cross_thread_implementation_of_memory_in/
2641 arxiv RESEARCH 0.180 Efficient Sequential Calibration with $O(T^{2/3-ε})$ Error Bound https://arxiv.org/abs/2607.12928v1
2166 rss UNCATEGORIZED 0.000 Google Images gets a Pinterest-like redesign focused on discovery https://techcrunch.com/2026/07/14/google-images-gets-a-pinterest-like-redesign-focused-on-discovery/
2642 arxiv RESEARCH 0.110 Knowledge- and Gradient-Guided Reinforcement Learning for Parametrized Action Markov Decision Processes https://arxiv.org/abs/2607.12924v1
2643 arxiv RESEARCH 0.180 LatentFlow: A General Framework for Conditioning Stochastic Processes https://arxiv.org/abs/2607.12922v1
2167 rss UNCATEGORIZED 0.000 AWS and Bluesight build AI for hospital 340B compliance https://www.artificialintelligence-news.com/news/aws-and-bluesight-build-ai-for-hospital-340b-compliance/
2315 reddit SHIPPING 0.460 Open Source Local LLM Training Tool (for consumer hardware) https://www.reddit.com/r/artificial/comments/1uwcah2/open_source_local_llm_training_tool_for_consumer/
2441 reddit PROBLEM SOLVED 0.330 New LLM Coordination Benchmark - Benchmarking Open-Ended Multi-Agent Coordination in Language Agents [R] https://www.reddit.com/r/MachineLearning/comments/1uwc6ni/new_llm_coordination_benchmark_benchmarking/
2326 reddit INFRASTRUCTURE 0.050 I'm not a great artist — so I made an agent that turns my doodles on my Remarkable tablet into actually nice charcoal sketches. Real editable pen-line vectors too! Not just static images. https://www.reddit.com/r/artificial/comments/1uwbt7o/im_not_a_great_artist_so_i_made_an_agent_that/
2219 hackernews UNCATEGORIZED 0.020 Are we offloading too much of our thinking to AI? https://www.artfish.ai/p/offloading-thinking-to-ai
2257 rss PROBLEM SOLVED 0.200 New York State halts construction of all new data centers https://techcrunch.com/2026/07/14/new-york-state-halts-construction-of-all-new-data-centers/
2325 reddit INFRASTRUCTURE 0.010 A new, state-of-the-art, agentic pipeline for easy Music Video creation https://www.reddit.com/r/artificial/comments/1uwbfos/a_new_stateoftheart_agentic_pipeline_for_easy/
2562 hackernews UNCATEGORIZED 0.020 The Agentic Loop: Three loops in a trench coat https://www.bobbytables.io/p/the-agentic-loop-three-loops-in-a
2253 rss BUSINESS 0.290 Reflection inks $1B compute deal with Nebius https://techcrunch.com/2026/07/14/reflection-inks-1b-compute-deal-with-nebius/
2254 rss CULTURE 0.290 The real AI race may no longer be at the frontier https://techcrunch.com/2026/07/14/the-real-ai-race-may-no-longer-be-at-the-frontier-open-models-hugging-face/
2255 rss UNCATEGORIZED 0.030 Spotify expands its AI push with a ChatGPT-like music assistant https://techcrunch.com/2026/07/14/spotify-expands-its-ai-push-with-a-chatgpt-like-music-assistant/
2256 rss UNCATEGORIZED 0.030 Superhuman’s new auto-draft feature almost makes me like AI replies https://techcrunch.com/2026/07/14/superhumans-new-auto-draft-feature-almost-makes-me-like-ai-replies/
2314 reddit INFRASTRUCTURE 0.190 The real bottleneck for AI agents may be proving who they are https://www.reddit.com/r/artificial/comments/1uw81un/the_real_bottleneck_for_ai_agents_may_be_proving/
2217 hackernews UNCATEGORIZED 0.020 Proof of care in the age of AI https://jacobfilipp.com/care/
2461 hackernews MODEL RELEASE 0.270 Show HN: I RL-trained an agent that trains models with RL (for ~$1.3k) https://github.com/Danau5tin/ai-trains-ai
2224 hackernews INFRASTRUCTURE 0.060 Coding agents think ahead of time https://arxiv.org/abs/2607.05188
2324 reddit BUSINESS 0.010 Did you know the CEO of OpenAI owns nearly 9% of Reddit while Reddit bans users for AI generated content? https://www.reddit.com/r/artificial/comments/1uw6sv6/did_you_know_the_ceo_of_openai_owns_nearly_9_of/
2258 rss CULTURE 0.000 ChatGPT returns to WhatsApp in Europe after EU forces Meta to open the door to rival AI bots https://the-decoder.com/chatgpt-returns-to-whatsapp-in-europe-after-eu-forces-meta-to-open-the-door-to-rival-ai-bots/
2260 rss PROBLEM SOLVED 0.240 Deepmind CEO Hassabis says "nobody in the world knows what happens next" so "cautious optimism" means building guardrails now https://the-decoder.com/deepmind-ceo-hassabis-says-nobody-in-the-world-knows-what-happens-next-so-cautious-optimism-means-building-guardrails-now/
2157 hackernews UNCATEGORIZED 0.140 Codex starts encrypting sub-agent prompts https://github.com/openai/codex/issues/28058
2262 rss BUSINESS 0.000 PixVerse's $2B valuation shows investors still believe AI video generation has room for another winner https://the-decoder.com/pixverses-2b-valuation-shows-investors-still-believe-ai-video-generation-has-room-for-another-winner/
2264 rss MODEL RELEASE 0.120 Claude responds with more warmth in Hindi and more rigor in Russian, showing how language shapes AI answers https://the-decoder.com/claude-values-study/
2220 hackernews UNCATEGORIZED 0.020 Demis Hassabis has a plan to harness AI safely https://twitter.com/demishassabis/status/2076957440109625718
2317 reddit UNCATEGORIZED 0.040 The first AI was a syllogism machine in 1956. We're still building the same thing. https://www.reddit.com/r/artificial/comments/1uw23qw/the_first_ai_was_a_syllogism_machine_in_1956_were/
2460 hackernews BUSINESS 0.020 OpenAI's Ad Business Is on Pace to Miss Its Own Forecast by 90%, Analyst Says https://www.adweek.com/media/openais-ad-business-is-on-pace-to-miss-its-own-forecast-by-90-analyst-says/
2432 reddit UNCATEGORIZED 0.050 How many on-the-fly augmentations per image for a single-class segmentation mode [R] https://www.reddit.com/r/MachineLearning/comments/1uvxt70/how_many_onthefly_augmentations_per_image_for_a/
2327 reddit UNCATEGORIZED 0.040 Inside Ghostcommit: How Malicious PNGs Bypass AI Code Reviewers https://www.reddit.com/r/artificial/comments/1uvxqg5/inside_ghostcommit_how_malicious_pngs_bypass_ai/
2152 reddit INFRASTRUCTURE 0.000 We keep asking whether AI will replace us. The more useful question is what it means to share the world with it. https://www.reddit.com/r/artificial/comments/1uvvd13/we_keep_asking_whether_ai_will_replace_us_the/
2266 rss UNCATEGORIZED 0.030 Uber’s product chief on hotels, robotaxis, and why the company doesn’t want to be ‘everything for everyone’ https://techcrunch.com/2026/07/13/ubers-product-chief-on-hotels-robotaxis-and-why-the-company-doesnt-want-to-be-everything-for-everyone/
2267 rss BUSINESS 0.000 Video-generation startup PixVerse raises $439M, valuation soars past $2B https://techcrunch.com/2026/07/13/video-generation-startup-pixverse-raises-439m-valuation-soars-past-2b/
2322 reddit MODEL RELEASE 0.100 Anthropic analyzed 300,000 real Claude conversations to measure its values. The findings are uncomfortable. https://www.reddit.com/r/artificial/comments/1uvpob7/anthropic_analyzed_300000_real_claude/
2158 hackernews UNCATEGORIZED 0.020 Samsung Health app threatens data deletion if users opt out AI training https://neow.in/cWsyMTV3
2226 hackernews UNCATEGORIZED 0.140 Show HN: I implemented a neural network in SQL https://github.com/xqlsystems/xarray-sql/blob/claude/xarray-sql-mnist-demo/benchmarks/nn.py
2221 hackernews UNCATEGORIZED 0.020 AI is a bad tool https://bytecode.news/posts/2026/07/user-submission-ai-is-a-bad-tool
2438 reddit PROBLEM SOLVED 0.410 GPUHedge: Hedging serverless GPU providers improves cold start p95 latency from 117s to 30s [P] https://www.reddit.com/r/MachineLearning/comments/1uvlb6h/gpuhedge_hedging_serverless_gpu_providers/
2223 hackernews INFRASTRUCTURE 0.060 Show HN: Nobie – an Excel-compatible runtime for agents and humans https://nobie.com
2126 rss CULTURE 0.000 The wildest allegations in Apple’s trade secrets lawsuit against OpenAI https://techcrunch.com/2026/07/13/the-wildest-allegations-in-apples-trade-secrets-lawsuit-against-openai/
2463 hackernews INFRASTRUCTURE 0.180 Show HN: BillAI Bass, an AI-Powered Big Mouth Billy Bass Using Strands Agents https://github.com/morganwilliscloud/billai-bass
2098 hackernews MODEL RELEASE 0.150 xAI's Grok Build CLI Uploads Git Repositories to a Google Cloud Bucket https://www.internationalcyberdigest.com/xais-grok-build-cli-uploads-entire-git-repositories-to-a-google-cloud-bucket/
2125 rss UNCATEGORIZED 0.000 What Anthropic’s latest AI discovery does—and doesn’t—show https://www.technologyreview.com/2026/07/13/1140343/what-anthropics-latest-ai-discovery-does-and-doesnt-show/
2144 arxiv RESEARCH 0.180 Requential Coding: Pushing the Limits of Model Compression with Self-Generated Training Data https://arxiv.org/abs/2607.11883v1
2145 arxiv RESEARCH 0.110 Metacognition in LLMs: Foundations, Progress, and Opportunities https://arxiv.org/abs/2607.11881v1
2146 arxiv RESEARCH 0.180 Invariant Learning Dynamics of Transformers in Inductive Reasoning Tasks https://arxiv.org/abs/2607.11875v1
2147 arxiv RESEARCH 0.300 A Minimalist Retargeting-Guided Reinforcement Learning Recipe for Dexterous Manipulation https://arxiv.org/abs/2607.11874v1
2148 arxiv RESEARCH 0.290 A Durability and Cross-Language Transfer Benchmark for a Validated Teaching-Feedback Classification Protocol https://arxiv.org/abs/2607.11873v1
2194 arxiv RESEARCH 0.150 Inside the Unfair Judge: A Mechanistic Interpretability Account of LLM-as-Judge Bias https://arxiv.org/abs/2607.11871v1
2444 reddit CULTURE 0.050 Chain of Thought is a scaling trap. the next wave is latent reasoning (Coconut / HRM / RecrusiveMAS)... but then we hit the black box wall. Where does BDH fit? [D] https://www.reddit.com/r/MachineLearning/comments/1uviru5/chain_of_thought_is_a_scaling_trap_the_next_wave/
2316 reddit INFRASTRUCTURE 0.050 The 'agent web' is coming — where AI agents talk directly to each other instead of scraping websites https://www.reddit.com/r/artificial/comments/1uviqvw/the_agent_web_is_coming_where_ai_agents_talk/
2195 arxiv RESEARCH 0.280 Evidence-Backed Video Question Answering https://arxiv.org/abs/2607.11862v1
2196 arxiv RESEARCH 0.280 AdvancedMathBench: A Benchmark Suite for Advanced Mathematical Proof Generation and Verification https://arxiv.org/abs/2607.11849v1
2197 arxiv RESEARCH 0.180 Input-Aware Dynamic Backdoor Attack Against Quantum Neural Networks https://arxiv.org/abs/2607.11843v1
2121 rss BUSINESS 0.000 Sam Altman’s space data center trash talk is what most experts already believe https://techcrunch.com/2026/07/13/sam-altmans-space-data-center-trash-talk-is-what-most-experts-already-believe/
2198 arxiv RESEARCH 0.110 LoRA-Based Cascaded Multimodal Fusion for Action Recognition in Medical Training Environments https://arxiv.org/abs/2607.11839v1
2199 arxiv RESEARCH 0.300 Transformer-Guided Swarm Intelligence for Frugal Neural Architecture Search https://arxiv.org/abs/2607.11826v1
2123 rss INFRASTRUCTURE 0.110 Turing Award winner Rich Sutton founds Oak Lab to build AI agents that learn on their own https://the-decoder.com/turing-award-winner-rich-sutton-founds-oak-lab-to-build-ai-agents-that-learn-on-their-own/
2200 arxiv RESEARCH 0.360 MM-ToolSandBox: A Unified Framework for Evaluating Visual Tool-Calling Agents https://arxiv.org/abs/2607.11818v1
2201 arxiv RESEARCH 0.150 Relaxing Faithfulness with Intervention-Only Causal Discovery https://arxiv.org/abs/2607.11816v1
2202 arxiv RESEARCH 0.110 Introducing Human-Centeredness in AI-Assisted Lexicography https://arxiv.org/abs/2607.11808v1
2203 arxiv RESEARCH 0.110 Encoder-Side Neuron Identification and Amplification for Acoustic Perception in Large Audio-Language Models https://arxiv.org/abs/2607.11801v1
2204 arxiv RESEARCH 0.140 StoryTeller: Training-Free Narrative Grounding for Long-Form Audio Description https://arxiv.org/abs/2607.11798v1
2205 arxiv RESEARCH 0.150 An Exact Instrument for State Usage in Selective State-Space Models, and the Input-Driven Migration It Reveals https://arxiv.org/abs/2607.11796v1
2318 reddit UNCATEGORIZED 0.010 Is there any kind of AI that could "read" huge loads of emails and give a "mark" according to a given expected result? https://www.reddit.com/r/artificial/comments/1uvgqrn/is_there_any_kind_of_ai_that_could_read_huge/
2206 arxiv RESEARCH 0.150 Forgetting Our Way to Shared Meaning: Effects of Forgetting on Conceptual Alignment in a Non-Partnership Coordination Game https://arxiv.org/abs/2607.11787v1
2207 arxiv RESEARCH 0.210 How Temperature Shapes Ideological Discourse in Retrieval-Augmented Generation? https://arxiv.org/abs/2607.11783v1
2130 rss UNCATEGORIZED 0.000 Should AI help you get away with killing your spouse? https://techcrunch.com/2026/07/13/should-ai-help-you-get-away-with-killing-your-spouse/
2208 arxiv RESEARCH 0.110 Evaluating RE Practices for Explainability: Synthesizing Insights from Daimler Truck into an Explainable RE Framework Proposal https://arxiv.org/abs/2607.11771v1
2129 rss UNCATEGORIZED 0.000 Nobel laureates and AI leaders warn the window to prepare for AI's economic impact is closing fast https://the-decoder.com/nobel-laureates-and-ai-leaders-warn-the-window-to-prepare-for-ais-economic-impact-is-closing-fast/
2222 hackernews UNCATEGORIZED 0.140 Show HN: Jacquard, a programming language for AI-written, human-reviewed code https://github.com/jbwinters/jacquard-lang
2131 rss MODEL RELEASE 0.090 Anthropic starts localizing Claude pricing for India, its biggest market after the US https://techcrunch.com/2026/07/13/anthropic-starts-localizing-claude-pricing-for-india-its-biggest-market-after-the-us/
1982 reddit LOCAL AI 0.330 Upgrade path for ryzen 9 (64 gb) + rtx 5080 https://www.reddit.com/r/LocalLLaMA/comments/1uvelii/upgrade_path_for_ryzen_9_64_gb_rtx_5080/
1972 reddit UNCATEGORIZED 0.010 Wan-Dancer: A Hierarchical Framework for Minute-scale Coherent Music-to-Dance Generation https://www.reddit.com/r/LocalLLaMA/comments/1uvdaq7/wandancer_a_hierarchical_framework_for/
2136 rss MODEL RELEASE 0.090 Nadella calls out AI labs like OpenAI and Anthropic for banning distillation while training on everyone else's data https://the-decoder.com/nadella-calls-out-ai-labs-like-openai-and-anthropic-for-banning-distillation-while-training-on-everyone-elses-data/
2127 rss MODEL RELEASE 0.120 Waze adds new AI-powered features and customization updates https://techcrunch.com/2026/07/13/waze-adds-new-ai-powered-features-and-customization-updates/
1969 reddit INFRASTRUCTURE 0.050 I benchmarked 15 "E-Waste" GPUs with Modern Workloads https://www.reddit.com/r/LocalLLaMA/comments/1uvcjd0/i_benchmarked_15_ewaste_gpus_with_modern_workloads/
2091 hackernews INFRASTRUCTURE 0.180 Show HN: Clawk – Give coding agents a disposable Linux VM, not your laptop https://github.com/clawkwork/clawk
2430 reddit SHIPPING 0.420 Hundreds of papers hit arXiv every day and maybe 3 matter to my research, so I built an open-source tool that finds them [P] https://www.reddit.com/r/MachineLearning/comments/1uvcdf7/hundreds_of_papers_hit_arxiv_every_day_and_maybe/
2093 hackernews MODEL RELEASE 0.110 Grok uploaded my user directory to xAI's servers https://twitter.com/a_green_being/status/2076598897779020159
1981 reddit UNCATEGORIZED 0.050 MCP…. Is bad? https://www.reddit.com/r/LocalLLaMA/comments/1uvaqxp/mcp_is_bad/
1973 reddit SHIPPING 0.350 Production Qwen 3.6-27B VLLM config? https://www.reddit.com/r/LocalLLaMA/comments/1uvacno/production_qwen_3627b_vllm_config/
2153 reddit PROBLEM SOLVED 0.120 I built a full 3D open-world racing game almost entirely with AI, and it now has real daily players. Here's the honest breakdown of what the model nailed and where it completely fell apart. https://www.reddit.com/r/artificial/comments/1uvaaf4/i_built_a_full_3d_openworld_racing_game_almost/
1971 reddit MODEL RELEASE 0.100 [Study/Models] Flint: Compressing Reasoning Without Breaking It https://www.reddit.com/r/LocalLLaMA/comments/1uv9o2u/studymodels_flint_compressing_reasoning_without/
2321 reddit CULTURE 0.000 Everyone keeps asking if AI will replace people. I think we’re asking the wrong question. https://www.reddit.com/r/artificial/comments/1uv9l8w/everyone_keeps_asking_if_ai_will_replace_people_i/
2133 rss MODEL RELEASE 0.190 German AI consortium releases Soofi S, an open 30B model that tops benchmarks in both English and German https://the-decoder.com/german-ai-consortium-releases-soofi-s-an-open-30b-model-that-tops-benchmarks-in-both-english-and-german/
2134 rss PROBLEM SOLVED 0.250 AI agent crawlers now need permission. Here’s how to get it https://www.artificialintelligence-news.com/news/ai-agent-crawlers-cloudflare-rules/
2035 rss MODEL RELEASE 0.090 Google’s SensorFM turns messy wearable sensor data into a general-purpose health intelligence layer https://the-decoder.com/sensorfm/
2149 reddit UNCATEGORIZED 0.040 For a silent revolution in the singularity scene https://www.reddit.com/r/artificial/comments/1uv63ms/for_a_silent_revolution_in_the_singularity_scene/
2084 hackernews UNCATEGORIZED 0.020 Zig Creator Calls Spade a Spade, Anthropic Blows Smoke https://raymyers.org/post/zed-creator-calls-spade-a-spade/
2433 reddit UNCATEGORIZED 0.050 Evaluating J-space entropy as an error predictor across 7 datasets on Qwen3-4B [R] https://www.reddit.com/r/MachineLearning/comments/1uv5l75/evaluating_jspace_entropy_as_an_error_predictor/
1970 reddit SHIPPING 0.280 Compressed Version of Qwen-3.6-27B coming from PrismML - Khosla-Backed Startup Claims Breakthrough With Largest-Ever AI Model on an iPhone https://www.reddit.com/r/LocalLLaMA/comments/1uv54fv/compressed_version_of_qwen3627b_coming_from/
2138 rss MODEL RELEASE 0.090 Anthropic extends free Fable 5 access for subscribers as OpenAI's GPT-5.6 Sol heats up the pricing war https://the-decoder.com/anthropic-extends-free-fable-5-access-for-subscribers-as-openais-gpt-5-6-sol-heats-up-the-pricing-war/
2319 reddit UNCATEGORIZED 0.010 The print success rates nobody talks about :Meshy vs Hi3D after 50+ models. https://www.reddit.com/r/artificial/comments/1uv50ty/the_print_success_rates_nobody_talks_about_meshy/
1979 reddit LOCAL AI 0.300 Experiment: autonomous NPCs powered by Gemma 4 E2B in the browser https://www.reddit.com/r/LocalLLaMA/comments/1uv3wnt/experiment_autonomous_npcs_powered_by_gemma_4_e2b/
2442 reddit PROBLEM SOLVED 0.260 Prompt-engineering paper accepted to ICML [R] https://www.reddit.com/r/MachineLearning/comments/1uv1xb3/promptengineering_paper_accepted_to_icml_r/
2313 reddit UNCATEGORIZED 0.010 Is the "J-Space" an emergent feature, or a strategic response to optimization pressure? https://www.reddit.com/r/artificial/comments/1uuz89v/is_the_jspace_an_emergent_feature_or_a_strategic/
2083 hackernews UNCATEGORIZED 0.020 Ask HN: Add flag for AI-generated articles https://news.ycombinator.com/item/48886741
2323 reddit INFRASTRUCTURE 0.050 AI agents may need an identity before they need more intelligence https://www.reddit.com/r/artificial/comments/1uuxhe6/ai_agents_may_need_an_identity_before_they_need/
1974 reddit PROBLEM SOLVED 0.180 Running Qwen3.5-122B on Mac Studio 96GB: Fixed 3 bugs that made long-context inference usable https://www.reddit.com/r/LocalLLaMA/comments/1uuwrc0/running_qwen35122b_on_mac_studio_96gb_fixed_3/
2151 reddit INFRASTRUCTURE 0.050 Someone built an AI agent that hacks networks and holds data for ransom. It just worked. https://www.reddit.com/r/artificial/comments/1uuouu7/someone_built_an_ai_agent_that_hacks_networks_and/
2097 hackernews UNCATEGORIZED 0.050 The One-Step Trap (In AI Research) http://incompleteideas.net/IncIdeas/OneStepTrap.html
2086 hackernews BUSINESS 0.000 I love LLMs, I hate hype https://geohot.github.io//blog/jekyll/update/2026/07/12/i-love-llms.html
2225 hackernews INFRASTRUCTURE 0.570 Show HN: Juggler – an open-source GUI coding agent, by the creator of JUCE https://github.com/juggler-ai/juggler
2095 hackernews INFRASTRUCTURE 0.020 Mechanistic interpretability researchers applying causality theory to LLMs https://cacm.acm.org/news/can-we-understand-how-large-language-models-reason/
2445 reddit PROBLEM SOLVED 0.260 Ph.D. in Operations Research / Big Tech Eng: How to transition into intermediate/advanced ML for high-value industries (Robotics, Defense, Finance)? [D] https://www.reddit.com/r/MachineLearning/comments/1uumkkg/phd_in_operations_research_big_tech_eng_how_to/
2089 hackernews PROBLEM SOLVED 0.410 Migrating a production AI agent to GPT-5.6: 2.2x faster, 27% cheaper https://ploy.ai/blog/migrating-a-production-ai-agent-to-gpt-5-6
2320 reddit UNCATEGORIZED 0.010 this openai court story is starting to look ugly https://www.reddit.com/r/artificial/comments/1uul5ef/this_openai_court_story_is_starting_to_look_ugly/
2039 rss UNCATEGORIZED 0.000 LinkedIn is the undisputed king of long-form AI slop, according to a study spanning five platforms https://the-decoder.com/linkedin-is-the-undisputed-king-of-long-form-ai-slop-according-to-a-study-spanning-five-platforms/
2037 rss MODEL RELEASE 0.090 Claude Code now has a built-in browser that lets the AI read, click, and type on external websites https://the-decoder.com/claude-code-now-has-a-built-in-browser-that-lets-the-ai-read-click-and-type-on-external-websites/
1571 reddit SHIPPING 0.190 Kreuzberg (local document extraction) is being renamed to Xberg - current version on LTS https://www.reddit.com/r/LocalLLaMA/comments/1uuhqlz/kreuzberg_local_document_extraction_is_being/
1980 reddit UNCATEGORIZED 0.170 Local Image to 3D (<2gb RAM, <20s, Apple Silicon, iPhone) https://www.reddit.com/r/LocalLLaMA/comments/1uuga40/local_image_to_3d_2gb_ram_20s_apple_silicon_iphone/
2092 hackernews PROBLEM SOLVED 0.190 AI boosts research careers but narrow the span of ideas explored: study https://spectrum.ieee.org/ai-science-research-flattens-discovery
1976 reddit LOCAL AI 0.440 If you use Open Code or other agenting programs you are leaving a lot of t/s if you don't actually use agents in parallel. Benchmark : RTX5090, Qwen3.6 35B loaded via LM studio with parallel tasks set to 8 https://www.reddit.com/r/LocalLLaMA/comments/1uueuks/if_you_use_open_code_or_other_agenting_programs/
2427 reddit SHIPPING 0.300 Zer0Fit: I took Google's new TabFM & TimesFM ML foundation models and made them available as an MCP server for zero-shot ML tasks (forecasts / classifications / regressions). 100% local. [P] https://www.reddit.com/r/MachineLearning/comments/1uue8cc/zer0fit_i_took_googles_new_tabfm_timesfm_ml/
1573 reddit LOCAL AI 0.260 I got Nemotron Puzzle 75B running smoothly on a 64GB M2 Max https://www.reddit.com/r/LocalLLaMA/comments/1uue46z/i_got_nemotron_puzzle_75b_running_smoothly_on_a/
1984 reddit UNCATEGORIZED 0.050 Working around Qwen3.6-27B's tool-call failures and looping https://www.reddit.com/r/LocalLLaMA/comments/1uue278/working_around_qwen3627bs_toolcall_failures_and/
1965 reddit SHIPPING 0.260 Zer0Fit: I took Google's new TabFM & TimesFM ML foundation models and made them available as an MCP server for zero-shot ML tasks (forecasts / classifications / regressions). 100% local. https://www.reddit.com/r/LocalLLaMA/comments/1uudxi8/zer0fit_i_took_googles_new_tabfm_timesfm_ml/
2041 rss PROBLEM SOLVED 0.170 S&P Global sees OpenAI as a "key credit risk" for Oracle and cuts its credit rating https://the-decoder.com/sp-global-sees-openai-as-a-key-credit-risk-for-oracle-and-cuts-its-credit-rating/
2431 reddit UNCATEGORIZED 0.050 Obtaining Irregular Learning Curves with HyberBand Tuned ANN model for Price Prediction [P] https://www.reddit.com/r/MachineLearning/comments/1uud3qj/obtaining_irregular_learning_curves_with/
2042 rss CULTURE 0.030 Meta kills Muse Image feature that let anyone generate AI photos of Instagram users without consent https://the-decoder.com/meta-kills-muse-image-feature-that-let-anyone-generate-ai-photos-of-instagram-users-without-consent/
2087 hackernews INFRASTRUCTURE 0.130 Old and new apps, via modern coding agents https://terrytao.wordpress.com/2026/07/11/old-and-new-apps-via-modern-coding-agents/
1968 reddit PROBLEM SOLVED 0.260 Benchmark - 4x 5060 Ti (64GB VRAM) (P2P) - Qwen3.6 27B (INT8 /w bf16 kv cache) @ 8 concurrency with SGLang. SGLang seems to handle higher concurrency better with this setup https://www.reddit.com/r/LocalLLaMA/comments/1uuc3pi/benchmark_4x_5060_ti_64gb_vram_p2p_qwen36_27b/
1923 rss MODEL RELEASE 0.090 Claude Cowork's biggest use case is the mundane office work nobody wants to own, Anthropic says https://the-decoder.com/claude-coworks-biggest-use-case-is-the-mundane-office-work-nobody-wants-to-own-anthropic-says/
1921 rss CULTURE 0.000 OpenAI CEO Altman is now "pretty sure" AI is net job-creating, which is quite the pivot from predicting mass layoffs https://the-decoder.com/openai-ceo-altman-is-now-pretty-sure-ai-is-net-job-creating-which-is-quite-the-pivot-from-predicting-mass-layoffs/
1977 reddit LOCAL AI 0.450 Voodoo Quant beats Unsloth Dynamic 2.0 KLD by 95% in Qwen3.5 0.8B and 2B https://www.reddit.com/r/LocalLLaMA/comments/1uua3jd/voodoo_quant_beats_unsloth_dynamic_20_kld_by_95/
1924 rss UNCATEGORIZED 0.000 Grades dropped from 96 to 48 percent when a Brown professor made students take the exam without AI https://the-decoder.com/grades-dropped-from-96-to-48-percent-when-a-brown-professor-made-students-take-the-exam-without-ai/
1824 rss INFRASTRUCTURE 0.040 AI agents win at Slay the Spire 2 after researchers replace growing chat logs with structured memory https://the-decoder.com/ai-agents-win-at-slay-the-spire-2-after-researchers-replace-growing-chat-logs-with-structured-memory/
1564 reddit MODEL RELEASE 0.100 Need help tuning cache in llama-server https://www.reddit.com/r/LocalLLaMA/comments/1uu8g9f/need_help_tuning_cache_in_llamaserver/
2094 hackernews INFRASTRUCTURE 0.180 Show HN: Mindwalk – Replay coding-agent sessions on a 3D map of your codebase https://github.com/cosmtrek/mindwalk
1569 reddit PROBLEM SOLVED 0.180 i would like to share my experience. working with huge LLMs and poor Machine https://www.reddit.com/r/LocalLLaMA/comments/1uu6qvh/i_would_like_to_share_my_experience_working_with/
1975 reddit INFRASTRUCTURE 0.320 **Your $80 Tesla P100 has been doing silently noisy math in llama.cpp for years. Three lines fix it, for free.** https://www.reddit.com/r/LocalLLaMA/comments/1uu6p9o/your_80_tesla_p100_has_been_doing_silently_noisy/
1978 reddit UNCATEGORIZED 0.010 I mapped Anthropic’s J-Space Hallucination signal across 7 datasets on Qwen3-4B to find out where it works and where it breaks https://www.reddit.com/r/LocalLLaMA/comments/1uu61wb/i_mapped_anthropics_jspace_hallucination_signal/
1566 reddit INFRASTRUCTURE 0.120 First attempts at a CPU setup - MS-02 Intel 285hx, trying Qwen3, Qwen3.6 and Gemma4 https://www.reddit.com/r/LocalLLaMA/comments/1uu5ht0/first_attempts_at_a_cpu_setup_ms02_intel_285hx/
1966 reddit UNCATEGORIZED 0.010 I didn't give up - extGemma4-40_5B returned https://www.reddit.com/r/LocalLLaMA/comments/1uu4hxp/i_didnt_give_up_extgemma440_5b_returned/
1572 reddit LOCAL AI 0.300 Qwenthropic https://www.reddit.com/r/LocalLLaMA/comments/1uu3545/qwenthropic/
1983 reddit LOCAL AI 0.300 Interactive Jacobian-Lens visualizer and live steerer for GGUF models on llama.cpp https://www.reddit.com/r/LocalLLaMA/comments/1uu32z6/interactive_jacobianlens_visualizer_and_live/
2085 hackernews MODEL RELEASE 0.270 What xAI's Grok build CLI sends to xAI: A wire-level analysis https://gist.github.com/cereblab/dc9a40bc26120f4540e4e09b75ffb547
1565 reddit LOCAL AI 0.340 Measuring PCIe transfer under dual GPU with pipeline & tensor llama.cpp https://www.reddit.com/r/LocalLLaMA/comments/1utz50z/measuring_pcie_transfer_under_dual_gpu_with/
2088 hackernews UNCATEGORIZED 0.020 Mesh LLM: distributed AI computing on iroh https://www.iroh.computer/blog/mesh-llm
2090 hackernews UNCATEGORIZED 0.020 Stop Telling Me to Ask an LLM https://blog.yaelwrites.com/stop-telling-me-to-ask-an-llm/
1967 reddit PROBLEM SOLVED 0.180 Ultra budget 20GB vram with 448GB/s for $100 bucks. https://www.reddit.com/r/LocalLLaMA/comments/1utwqf8/ultra_budget_20gb_vram_with_448gbs_for_100_bucks/
1559 reddit INFRASTRUCTURE 0.340 Performance comparison on full compute performance (Anima) and LLM prompt processing of 5090 (600,475 and 400W) vs 6000 PRO MaxQ shunt modded and water cooled (at 300, 400, 475 and 600W), and 6000 PRO WS/SE (600W). https://www.reddit.com/r/LocalLLaMA/comments/1utvbey/performance_comparison_on_full_compute/
1558 reddit LOCAL AI 0.500 I benched quad 5060Tis for code generation with Qwen3.6-27B so you don't have to (it's really good) https://www.reddit.com/r/LocalLLaMA/comments/1uturng/i_benched_quad_5060tis_for_code_generation_with/
2096 hackernews BUSINESS 0.020 Wealthy AI workers send San Francisco house prices soaring https://www.bbc.com/news/articles/c9q29j47v9ro
1992 hackernews CULTURE 0.020 AI 2040 and the cult of intelligence https://geohot.github.io//blog/jekyll/update/2026/07/11/ai-2040.html
2000 hackernews INFRASTRUCTURE 0.100 Who manages the agents? https://www.off-policy.com/dont-go-quietly-into-the-ai-night/
1828 rss UNCATEGORIZED 0.000 OpenAI's GPT-5.6 Sol Ultra reportedly solves a 50-year-old math problem in under an hour https://the-decoder.com/openais-gpt-5-6-sol-ultra-reportedly-solves-a-50-year-old-math-problem-in-under-an-hour/
1999 hackernews CULTURE 0.020 Reverse centaurs are the answer to the AI paradox (2025) https://pluralistic.net/2025/09/11/vulgar-thatcherism/#there-is-an-alternative
1731 rss MODEL RELEASE 0.090 Terrorist groups are using every major AI chatbot for attack planning and weapons development https://the-decoder.com/terrorist-groups-are-using-every-major-ai-chatbot-for-attack-planning-and-weapons-development/
2001 hackernews INFRASTRUCTURE 0.180 Show HN: Reame – a CPU inference server that gets faster as it runs https://github.com/swellweb/reame
1730 rss BUSINESS 0.000 OpenAI bets on families as ChatGPT goes deeper into households https://techcrunch.com/2026/07/11/openai-bets-on-families-as-chatgpt-goes-deeper-into-households/
1776 hackernews UNCATEGORIZED 0.020 Ghost Font: A font that humans can read but AI cannot https://www.mixfont.com/ghost-font
1786 hackernews BUSINESS 0.020 Microsoft latest report shows 25% emissions raised due to AI data centers https://www.windowscentral.com/microsoft/dropping-greenwashing-credits-and-expanding-ai-datacenters-caused-microsofts-25-percent-emissions-jump
1592 hackernews PROBLEM SOLVED 0.190 Companies are scrambling to curtail soaring AI costs https://www.economist.com/business/2026/06/14/companies-are-scrambling-to-curtail-soaring-ai-costs
1591 hackernews SHIPPING 0.230 Meta pulls new AI image feature after days of backlash https://www.bbc.com/news/articles/c2dy6e8klw0o
1590 hackernews PROBLEM SOLVED 0.190 AI Can't Recreate the Thrust Game (But It Can Help You Understand It) https://www.jamesdrandall.com/posts/thrust_ai_powered_software_archaeology/
1588 hackernews CULTURE 0.020 Apple sues OpenAI, accusing it of stealing company secrets https://www.nytimes.com/2026/07/10/technology/apple-openai-lawsuit.html
athena_review.tsv
+201
View File
@@ -0,0 +1,201 @@
id source bucket final_score verdict title url
2702 rss SHIPPING 0.180 OpenAI researcher Miles Wang in talks to launch AI drug discovery startup valued at $2B https://techcrunch.com/2026/07/14/openai-researcher-miles-wang-in-talks-to-launch-ai-drug-discovery-startup-valued-at-2b/
2587 rss CULTURE 0.000 Lorde says AI glasses are ‘not sexy’ https://techcrunch.com/2026/07/14/lorde-says-ai-glasses-are-not-sexy/
2584 rss UNCATEGORIZED 0.030 OpenAI’s first hardware device is reportedly a screenless speaker that can move https://techcrunch.com/2026/07/14/openais-first-hardware-device-is-reportedly-a-screenless-speaker-that-can-move/
2589 rss CULTURE 0.000 OpenAI pushes back on Apple trade secret lawsuit https://techcrunch.com/2026/07/14/openai-pushes-back-on-apple-trade-secret-lawsuit/
2672 hackernews UNCATEGORIZED 0.020 Financing the AI boom: from cash flows to debt [pdf] https://www.bis.org/publ/bisbull120.pdf
2585 rss MODEL RELEASE 0.120 OpenAI’s new flagship model deletes files on its own, people keep warning https://techcrunch.com/2026/07/14/openais-new-flagship-model-deletes-files-on-its-own-people-keep-warning/
2646 reddit MODEL RELEASE 0.130 Opening the Black Box: Unison Zero Parameter Model https://www.reddit.com/r/artificial/comments/1uwjwl6/opening_the_black_box_unison_zero_parameter_model/
2648 reddit PROBLEM SOLVED 0.220 Developers Hate AI. I Used It To Sell 10 Websites This Week. https://www.reddit.com/r/artificial/comments/1uwj75g/developers_hate_ai_i_used_it_to_sell_10_websites/
2485 rss SHIPPING 0.210 Apple opens its new Siri AI to everyone with the iOS 27 public beta https://techcrunch.com/2026/07/14/apple-opens-its-new-siri-ai-to-everyone-with-the-ios-27-public-beta/
2486 rss CULTURE 0.000 Anthropic’s newest ad is creeping people out https://techcrunch.com/2026/07/14/anthropics-newest-ad-is-creeping-people-out/
2487 rss BUSINESS 0.030 The founder of Hinge raised $18M to build a new AI dating service, Overtone https://techcrunch.com/2026/07/14/the-founder-of-hinge-raised-18m-to-build-a-new-ai-dating-service-overtone/
2654 reddit BUSINESS 0.010 How does a 102M-parameter transformer forecast multivariate time series? https://www.reddit.com/r/artificial/comments/1uwh9ko/how_does_a_102mparameter_transformer_forecast/
2490 rss MODEL RELEASE 0.090 Google faces another AI training lawsuit from major publishers https://techcrunch.com/2026/07/14/google-faces-another-ai-training-lawsuit-from-major-publishers/
2652 reddit CULTURE 0.000 Apple just sued OpenAI for trade secret theft. And Google quietly rewrote how the internet works. https://www.reddit.com/r/artificial/comments/1uwh06x/apple_just_sued_openai_for_trade_secret_theft_and/
2645 reddit SHIPPING 0.230 The absolute nightmare of putting AI agents into actual production https://www.reddit.com/r/artificial/comments/1uwg8kk/the_absolute_nightmare_of_putting_ai_agents_into/
2624 arxiv RESEARCH 0.190 Do AI Agents Know When a Task Is Simple? Toward Complexity-Aware Reasoning and Execution https://arxiv.org/abs/2607.13034v1
2625 arxiv RESEARCH 0.150 The Seriality Gap in Video Diffusion Models https://arxiv.org/abs/2607.13031v1
2626 arxiv RESEARCH 0.340 TerraZero: Procedural Driving Simulation for Zero-Demonstration Self-Play at Scale https://arxiv.org/abs/2607.13028v1
2627 arxiv RESEARCH 0.340 PalmClaw: A Native On-Device Agent Framework for Mobile Phones https://arxiv.org/abs/2607.13027v1
2628 arxiv RESEARCH 0.150 A Shortcut to Statistically Steady-State Turbulence with Flow Matching https://arxiv.org/abs/2607.13022v1
2649 reddit PROBLEM SOLVED 0.210 Ford replaced engineers with AI, then quietly hired 350 back. The reason should stop every founder about to cut their team to SAVE money. https://www.reddit.com/r/artificial/comments/1uwg31g/ford_replaced_engineers_with_ai_then_quietly/
2629 arxiv RESEARCH 0.110 Audio-Native Speech Recognition with a Frozen Discrete-Diffusion Language Model https://arxiv.org/abs/2607.13013v1
2630 arxiv RESEARCH 0.140 Dynamic Resource Allocation for Ensemble Determinization MCTS https://arxiv.org/abs/2607.13007v1
2631 arxiv RESEARCH 0.180 The Spectrum Is Not Enough: When Context Helps Time-Series Forecasting https://arxiv.org/abs/2607.13006v1
2632 arxiv RESEARCH 0.150 Watermark Forensics for Generative Models: An Information-Theoretic Perspective https://arxiv.org/abs/2607.13003v1
2494 rss MODEL RELEASE 0.090 DeepMind CEO calls for an independent standards body to regulate frontier AI https://techcrunch.com/2026/07/14/deepmind-ceo-calls-for-an-independent-standards-body-to-regulate-frontier-ai/
2429 reddit MODEL RELEASE 0.210 [P] RL-training Qwen3.6 to RL-train tool using AI models [P] https://www.reddit.com/r/MachineLearning/comments/1uwfmfa/p_rltraining_qwen36_to_rltrain_tool_using_ai/
2633 arxiv RESEARCH 0.110 Win by Silence: Deletion Non-Monotonicity, Autonomous Exploitation, and Typed-State Gating in LLM Plan Evaluation https://arxiv.org/abs/2607.12986v1
2634 arxiv RESEARCH 0.110 Resist and Update: Counterfactual Report Coordinates for Incentive-Compatible LLMs https://arxiv.org/abs/2607.12985v1
2635 arxiv RESEARCH 0.140 FormalAnalyticGeo: A Neural-Symbolic Based Framework for Multimodal Analytic Geometry Problem Generation https://arxiv.org/abs/2607.12982v1
2488 rss MODEL RELEASE 0.090 Anthropic opens Claude for Teachers with a promise not to train models on student data https://the-decoder.com/anthropic-opens-claude-for-teachers-with-a-promise-not-to-train-models-on-student-data/
2636 arxiv RESEARCH 0.180 Ensemble Controlled-Flow Filtering for Implicit Data Assimilation https://arxiv.org/abs/2607.12975v1
2637 arxiv RESEARCH 0.210 The Illusion of Robustness: Aggregate Accuracy Hides Prediction Flips under Task-Irrelevant Context https://arxiv.org/abs/2607.12963v1
2638 arxiv RESEARCH 0.440 Form, Not Content? A Preregistered, Placebo-Controlled Evaluation of Learned Error-Conditioned Self-Repair Through Prompts and Weights in Frozen Small Code Models https://arxiv.org/abs/2607.12962v1
2656 reddit SHIPPING 0.190 Structured output reliability with LLMs — 3-month production learnings https://www.reddit.com/r/artificial/comments/1uwe9qp/structured_output_reliability_with_llms_3month/
2639 arxiv RESEARCH 0.190 Robustness of Deep Learning Models for PV Power Forecasting under NWP Forecast Errors: A Spatiotemporal and Physically Interpretable Analysis https://arxiv.org/abs/2607.12954v1
2640 arxiv RESEARCH 0.250 ViHoRec: A Quality-Controlled Vietnamese Hotel Recommendation Dataset and Cold-Start Benchmark https://arxiv.org/abs/2607.12946v1
2502 rss BUSINESS 0.030 DeepSeek needs more cash just weeks after closing its first $7 billion round https://the-decoder.com/deepseek-needs-more-cash-just-weeks-after-closing-its-first-7-billion-round/
2168 rss CULTURE 0.000 Meta’s Adam Mosseri says AI token budgets could soon be capped per engineer https://techcrunch.com/2026/07/14/metas-adam-mosseri-says-ai-token-budgets-could-soon-be-capped-per-engineer/
2164 rss UNCATEGORIZED 0.030 Google Search now generates AI images when it can't find what you're looking for on the web https://the-decoder.com/google-search-now-generates-ai-images-when-it-cant-find-what-youre-looking-for-on-the-web/
2150 reddit MODEL RELEASE 0.140 All cross thread implementation of memory in chatgpt, claude, and gemini is unsafe https://www.reddit.com/r/artificial/comments/1uwdc0k/all_cross_thread_implementation_of_memory_in/
2641 arxiv RESEARCH 0.180 Efficient Sequential Calibration with $O(T^{2/3-ε})$ Error Bound https://arxiv.org/abs/2607.12928v1
2166 rss UNCATEGORIZED 0.000 Google Images gets a Pinterest-like redesign focused on discovery https://techcrunch.com/2026/07/14/google-images-gets-a-pinterest-like-redesign-focused-on-discovery/
2642 arxiv RESEARCH 0.110 Knowledge- and Gradient-Guided Reinforcement Learning for Parametrized Action Markov Decision Processes https://arxiv.org/abs/2607.12924v1
2643 arxiv RESEARCH 0.180 LatentFlow: A General Framework for Conditioning Stochastic Processes https://arxiv.org/abs/2607.12922v1
2167 rss UNCATEGORIZED 0.000 AWS and Bluesight build AI for hospital 340B compliance https://www.artificialintelligence-news.com/news/aws-and-bluesight-build-ai-for-hospital-340b-compliance/
2315 reddit SHIPPING 0.460 Open Source Local LLM Training Tool (for consumer hardware) https://www.reddit.com/r/artificial/comments/1uwcah2/open_source_local_llm_training_tool_for_consumer/
2441 reddit PROBLEM SOLVED 0.330 New LLM Coordination Benchmark - Benchmarking Open-Ended Multi-Agent Coordination in Language Agents [R] https://www.reddit.com/r/MachineLearning/comments/1uwc6ni/new_llm_coordination_benchmark_benchmarking/
2326 reddit INFRASTRUCTURE 0.050 I'm not a great artist — so I made an agent that turns my doodles on my Remarkable tablet into actually nice charcoal sketches. Real editable pen-line vectors too! Not just static images. https://www.reddit.com/r/artificial/comments/1uwbt7o/im_not_a_great_artist_so_i_made_an_agent_that/
2219 hackernews UNCATEGORIZED 0.020 Are we offloading too much of our thinking to AI? https://www.artfish.ai/p/offloading-thinking-to-ai
2257 rss PROBLEM SOLVED 0.200 New York State halts construction of all new data centers https://techcrunch.com/2026/07/14/new-york-state-halts-construction-of-all-new-data-centers/
2325 reddit INFRASTRUCTURE 0.010 A new, state-of-the-art, agentic pipeline for easy Music Video creation https://www.reddit.com/r/artificial/comments/1uwbfos/a_new_stateoftheart_agentic_pipeline_for_easy/
2562 hackernews UNCATEGORIZED 0.020 The Agentic Loop: Three loops in a trench coat https://www.bobbytables.io/p/the-agentic-loop-three-loops-in-a
2253 rss BUSINESS 0.290 Reflection inks $1B compute deal with Nebius https://techcrunch.com/2026/07/14/reflection-inks-1b-compute-deal-with-nebius/
2254 rss CULTURE 0.290 The real AI race may no longer be at the frontier https://techcrunch.com/2026/07/14/the-real-ai-race-may-no-longer-be-at-the-frontier-open-models-hugging-face/
2255 rss UNCATEGORIZED 0.030 Spotify expands its AI push with a ChatGPT-like music assistant https://techcrunch.com/2026/07/14/spotify-expands-its-ai-push-with-a-chatgpt-like-music-assistant/
2256 rss UNCATEGORIZED 0.030 Superhuman’s new auto-draft feature almost makes me like AI replies https://techcrunch.com/2026/07/14/superhumans-new-auto-draft-feature-almost-makes-me-like-ai-replies/
2314 reddit INFRASTRUCTURE 0.190 The real bottleneck for AI agents may be proving who they are https://www.reddit.com/r/artificial/comments/1uw81un/the_real_bottleneck_for_ai_agents_may_be_proving/
2217 hackernews UNCATEGORIZED 0.020 Proof of care in the age of AI https://jacobfilipp.com/care/
2461 hackernews MODEL RELEASE 0.270 Show HN: I RL-trained an agent that trains models with RL (for ~$1.3k) https://github.com/Danau5tin/ai-trains-ai
2224 hackernews INFRASTRUCTURE 0.060 Coding agents think ahead of time https://arxiv.org/abs/2607.05188
2324 reddit BUSINESS 0.010 Did you know the CEO of OpenAI owns nearly 9% of Reddit while Reddit bans users for AI generated content? https://www.reddit.com/r/artificial/comments/1uw6sv6/did_you_know_the_ceo_of_openai_owns_nearly_9_of/
2258 rss CULTURE 0.000 ChatGPT returns to WhatsApp in Europe after EU forces Meta to open the door to rival AI bots https://the-decoder.com/chatgpt-returns-to-whatsapp-in-europe-after-eu-forces-meta-to-open-the-door-to-rival-ai-bots/
2260 rss PROBLEM SOLVED 0.240 "Deepmind CEO Hassabis says ""nobody in the world knows what happens next"" so ""cautious optimism"" means building guardrails now" https://the-decoder.com/deepmind-ceo-hassabis-says-nobody-in-the-world-knows-what-happens-next-so-cautious-optimism-means-building-guardrails-now/
2157 hackernews UNCATEGORIZED 0.140 Codex starts encrypting sub-agent prompts https://github.com/openai/codex/issues/28058
2262 rss BUSINESS 0.000 PixVerse's $2B valuation shows investors still believe AI video generation has room for another winner https://the-decoder.com/pixverses-2b-valuation-shows-investors-still-believe-ai-video-generation-has-room-for-another-winner/
2264 rss MODEL RELEASE 0.120 Claude responds with more warmth in Hindi and more rigor in Russian, showing how language shapes AI answers https://the-decoder.com/claude-values-study/
2220 hackernews UNCATEGORIZED 0.020 Demis Hassabis has a plan to harness AI safely https://twitter.com/demishassabis/status/2076957440109625718
2317 reddit UNCATEGORIZED 0.040 The first AI was a syllogism machine in 1956. We're still building the same thing. https://www.reddit.com/r/artificial/comments/1uw23qw/the_first_ai_was_a_syllogism_machine_in_1956_were/
2460 hackernews BUSINESS 0.020 OpenAI's Ad Business Is on Pace to Miss Its Own Forecast by 90%, Analyst Says https://www.adweek.com/media/openais-ad-business-is-on-pace-to-miss-its-own-forecast-by-90-analyst-says/
2432 reddit UNCATEGORIZED 0.050 How many on-the-fly augmentations per image for a single-class segmentation mode [R] https://www.reddit.com/r/MachineLearning/comments/1uvxt70/how_many_onthefly_augmentations_per_image_for_a/
2327 reddit UNCATEGORIZED 0.040 Inside Ghostcommit: How Malicious PNGs Bypass AI Code Reviewers https://www.reddit.com/r/artificial/comments/1uvxqg5/inside_ghostcommit_how_malicious_pngs_bypass_ai/
2152 reddit INFRASTRUCTURE 0.000 We keep asking whether AI will replace us. The more useful question is what it means to share the world with it. https://www.reddit.com/r/artificial/comments/1uvvd13/we_keep_asking_whether_ai_will_replace_us_the/
2266 rss UNCATEGORIZED 0.030 Uber’s product chief on hotels, robotaxis, and why the company doesn’t want to be ‘everything for everyone’ https://techcrunch.com/2026/07/13/ubers-product-chief-on-hotels-robotaxis-and-why-the-company-doesnt-want-to-be-everything-for-everyone/
2267 rss BUSINESS 0.000 Video-generation startup PixVerse raises $439M, valuation soars past $2B https://techcrunch.com/2026/07/13/video-generation-startup-pixverse-raises-439m-valuation-soars-past-2b/
2322 reddit MODEL RELEASE 0.100 Anthropic analyzed 300,000 real Claude conversations to measure its values. The findings are uncomfortable. https://www.reddit.com/r/artificial/comments/1uvpob7/anthropic_analyzed_300000_real_claude/
2158 hackernews UNCATEGORIZED 0.020 Samsung Health app threatens data deletion if users opt out AI training https://neow.in/cWsyMTV3
2226 hackernews UNCATEGORIZED 0.140 Show HN: I implemented a neural network in SQL https://github.com/xqlsystems/xarray-sql/blob/claude/xarray-sql-mnist-demo/benchmarks/nn.py
2221 hackernews UNCATEGORIZED 0.020 AI is a bad tool https://bytecode.news/posts/2026/07/user-submission-ai-is-a-bad-tool
2438 reddit PROBLEM SOLVED 0.410 GPUHedge: Hedging serverless GPU providers improves cold start p95 latency from 117s to 30s [P] https://www.reddit.com/r/MachineLearning/comments/1uvlb6h/gpuhedge_hedging_serverless_gpu_providers/
2223 hackernews INFRASTRUCTURE 0.060 Show HN: Nobie – an Excel-compatible runtime for agents and humans https://nobie.com
2126 rss CULTURE 0.000 The wildest allegations in Apple’s trade secrets lawsuit against OpenAI https://techcrunch.com/2026/07/13/the-wildest-allegations-in-apples-trade-secrets-lawsuit-against-openai/
2463 hackernews INFRASTRUCTURE 0.180 Show HN: BillAI Bass, an AI-Powered Big Mouth Billy Bass Using Strands Agents https://github.com/morganwilliscloud/billai-bass
2098 hackernews MODEL RELEASE 0.150 xAI's Grok Build CLI Uploads Git Repositories to a Google Cloud Bucket https://www.internationalcyberdigest.com/xais-grok-build-cli-uploads-entire-git-repositories-to-a-google-cloud-bucket/
2125 rss UNCATEGORIZED 0.000 What Anthropic’s latest AI discovery does—and doesn’t—show https://www.technologyreview.com/2026/07/13/1140343/what-anthropics-latest-ai-discovery-does-and-doesnt-show/
2144 arxiv RESEARCH 0.180 Requential Coding: Pushing the Limits of Model Compression with Self-Generated Training Data https://arxiv.org/abs/2607.11883v1
2145 arxiv RESEARCH 0.110 Metacognition in LLMs: Foundations, Progress, and Opportunities https://arxiv.org/abs/2607.11881v1
2146 arxiv RESEARCH 0.180 Invariant Learning Dynamics of Transformers in Inductive Reasoning Tasks https://arxiv.org/abs/2607.11875v1
2147 arxiv RESEARCH 0.300 A Minimalist Retargeting-Guided Reinforcement Learning Recipe for Dexterous Manipulation https://arxiv.org/abs/2607.11874v1
2148 arxiv RESEARCH 0.290 A Durability and Cross-Language Transfer Benchmark for a Validated Teaching-Feedback Classification Protocol https://arxiv.org/abs/2607.11873v1
2194 arxiv RESEARCH 0.150 Inside the Unfair Judge: A Mechanistic Interpretability Account of LLM-as-Judge Bias https://arxiv.org/abs/2607.11871v1
2444 reddit CULTURE 0.050 Chain of Thought is a scaling trap. the next wave is latent reasoning (Coconut / HRM / RecrusiveMAS)... but then we hit the black box wall. Where does BDH fit? [D] https://www.reddit.com/r/MachineLearning/comments/1uviru5/chain_of_thought_is_a_scaling_trap_the_next_wave/
2316 reddit INFRASTRUCTURE 0.050 The 'agent web' is coming — where AI agents talk directly to each other instead of scraping websites https://www.reddit.com/r/artificial/comments/1uviqvw/the_agent_web_is_coming_where_ai_agents_talk/
2195 arxiv RESEARCH 0.280 Evidence-Backed Video Question Answering https://arxiv.org/abs/2607.11862v1
2196 arxiv RESEARCH 0.280 AdvancedMathBench: A Benchmark Suite for Advanced Mathematical Proof Generation and Verification https://arxiv.org/abs/2607.11849v1
2197 arxiv RESEARCH 0.180 Input-Aware Dynamic Backdoor Attack Against Quantum Neural Networks https://arxiv.org/abs/2607.11843v1
2121 rss BUSINESS 0.000 Sam Altman’s space data center trash talk is what most experts already believe https://techcrunch.com/2026/07/13/sam-altmans-space-data-center-trash-talk-is-what-most-experts-already-believe/
2198 arxiv RESEARCH 0.110 LoRA-Based Cascaded Multimodal Fusion for Action Recognition in Medical Training Environments https://arxiv.org/abs/2607.11839v1
2199 arxiv RESEARCH 0.300 Transformer-Guided Swarm Intelligence for Frugal Neural Architecture Search https://arxiv.org/abs/2607.11826v1
2123 rss INFRASTRUCTURE 0.110 Turing Award winner Rich Sutton founds Oak Lab to build AI agents that learn on their own https://the-decoder.com/turing-award-winner-rich-sutton-founds-oak-lab-to-build-ai-agents-that-learn-on-their-own/
2200 arxiv RESEARCH 0.360 MM-ToolSandBox: A Unified Framework for Evaluating Visual Tool-Calling Agents https://arxiv.org/abs/2607.11818v1
2201 arxiv RESEARCH 0.150 Relaxing Faithfulness with Intervention-Only Causal Discovery https://arxiv.org/abs/2607.11816v1
2202 arxiv RESEARCH 0.110 Introducing Human-Centeredness in AI-Assisted Lexicography https://arxiv.org/abs/2607.11808v1
2203 arxiv RESEARCH 0.110 Encoder-Side Neuron Identification and Amplification for Acoustic Perception in Large Audio-Language Models https://arxiv.org/abs/2607.11801v1
2204 arxiv RESEARCH 0.140 StoryTeller: Training-Free Narrative Grounding for Long-Form Audio Description https://arxiv.org/abs/2607.11798v1
2205 arxiv RESEARCH 0.150 An Exact Instrument for State Usage in Selective State-Space Models, and the Input-Driven Migration It Reveals https://arxiv.org/abs/2607.11796v1
2318 reddit UNCATEGORIZED 0.010 "Is there any kind of AI that could ""read"" huge loads of emails and give a ""mark"" according to a given expected result?" https://www.reddit.com/r/artificial/comments/1uvgqrn/is_there_any_kind_of_ai_that_could_read_huge/
2206 arxiv RESEARCH 0.150 Forgetting Our Way to Shared Meaning: Effects of Forgetting on Conceptual Alignment in a Non-Partnership Coordination Game https://arxiv.org/abs/2607.11787v1
2207 arxiv RESEARCH 0.210 How Temperature Shapes Ideological Discourse in Retrieval-Augmented Generation? https://arxiv.org/abs/2607.11783v1
2130 rss UNCATEGORIZED 0.000 Should AI help you get away with killing your spouse? https://techcrunch.com/2026/07/13/should-ai-help-you-get-away-with-killing-your-spouse/
2208 arxiv RESEARCH 0.110 Evaluating RE Practices for Explainability: Synthesizing Insights from Daimler Truck into an Explainable RE Framework Proposal https://arxiv.org/abs/2607.11771v1
2129 rss UNCATEGORIZED 0.000 Nobel laureates and AI leaders warn the window to prepare for AI's economic impact is closing fast https://the-decoder.com/nobel-laureates-and-ai-leaders-warn-the-window-to-prepare-for-ais-economic-impact-is-closing-fast/
2222 hackernews UNCATEGORIZED 0.140 Show HN: Jacquard, a programming language for AI-written, human-reviewed code https://github.com/jbwinters/jacquard-lang
2131 rss MODEL RELEASE 0.090 Anthropic starts localizing Claude pricing for India, its biggest market after the US https://techcrunch.com/2026/07/13/anthropic-starts-localizing-claude-pricing-for-india-its-biggest-market-after-the-us/
1982 reddit LOCAL AI 0.330 Upgrade path for ryzen 9 (64 gb) + rtx 5080 https://www.reddit.com/r/LocalLLaMA/comments/1uvelii/upgrade_path_for_ryzen_9_64_gb_rtx_5080/
1972 reddit UNCATEGORIZED 0.010 Wan-Dancer: A Hierarchical Framework for Minute-scale Coherent Music-to-Dance Generation https://www.reddit.com/r/LocalLLaMA/comments/1uvdaq7/wandancer_a_hierarchical_framework_for/
2136 rss MODEL RELEASE 0.090 Nadella calls out AI labs like OpenAI and Anthropic for banning distillation while training on everyone else's data https://the-decoder.com/nadella-calls-out-ai-labs-like-openai-and-anthropic-for-banning-distillation-while-training-on-everyone-elses-data/
2127 rss MODEL RELEASE 0.120 Waze adds new AI-powered features and customization updates https://techcrunch.com/2026/07/13/waze-adds-new-ai-powered-features-and-customization-updates/
1969 reddit INFRASTRUCTURE 0.050 "I benchmarked 15 ""E-Waste"" GPUs with Modern Workloads" https://www.reddit.com/r/LocalLLaMA/comments/1uvcjd0/i_benchmarked_15_ewaste_gpus_with_modern_workloads/
2091 hackernews INFRASTRUCTURE 0.180 Show HN: Clawk – Give coding agents a disposable Linux VM, not your laptop https://github.com/clawkwork/clawk
2430 reddit SHIPPING 0.420 Hundreds of papers hit arXiv every day and maybe 3 matter to my research, so I built an open-source tool that finds them [P] https://www.reddit.com/r/MachineLearning/comments/1uvcdf7/hundreds_of_papers_hit_arxiv_every_day_and_maybe/
2093 hackernews MODEL RELEASE 0.110 Grok uploaded my user directory to xAI's servers https://twitter.com/a_green_being/status/2076598897779020159
1981 reddit UNCATEGORIZED 0.050 MCP…. Is bad? https://www.reddit.com/r/LocalLLaMA/comments/1uvaqxp/mcp_is_bad/
1973 reddit SHIPPING 0.350 Production Qwen 3.6-27B VLLM config? https://www.reddit.com/r/LocalLLaMA/comments/1uvacno/production_qwen_3627b_vllm_config/
2153 reddit PROBLEM SOLVED 0.120 I built a full 3D open-world racing game almost entirely with AI, and it now has real daily players. Here's the honest breakdown of what the model nailed and where it completely fell apart. https://www.reddit.com/r/artificial/comments/1uvaaf4/i_built_a_full_3d_openworld_racing_game_almost/
1971 reddit MODEL RELEASE 0.100 [Study/Models] Flint: Compressing Reasoning Without Breaking It https://www.reddit.com/r/LocalLLaMA/comments/1uv9o2u/studymodels_flint_compressing_reasoning_without/
2321 reddit CULTURE 0.000 Everyone keeps asking if AI will replace people. I think we’re asking the wrong question. https://www.reddit.com/r/artificial/comments/1uv9l8w/everyone_keeps_asking_if_ai_will_replace_people_i/
2133 rss MODEL RELEASE 0.190 German AI consortium releases Soofi S, an open 30B model that tops benchmarks in both English and German https://the-decoder.com/german-ai-consortium-releases-soofi-s-an-open-30b-model-that-tops-benchmarks-in-both-english-and-german/
2134 rss PROBLEM SOLVED 0.250 AI agent crawlers now need permission. Here’s how to get it https://www.artificialintelligence-news.com/news/ai-agent-crawlers-cloudflare-rules/
2035 rss MODEL RELEASE 0.090 Google’s SensorFM turns messy wearable sensor data into a general-purpose health intelligence layer https://the-decoder.com/sensorfm/
2149 reddit UNCATEGORIZED 0.040 For a silent revolution in the singularity scene https://www.reddit.com/r/artificial/comments/1uv63ms/for_a_silent_revolution_in_the_singularity_scene/
2084 hackernews UNCATEGORIZED 0.020 Zig Creator Calls Spade a Spade, Anthropic Blows Smoke https://raymyers.org/post/zed-creator-calls-spade-a-spade/
2433 reddit UNCATEGORIZED 0.050 Evaluating J-space entropy as an error predictor across 7 datasets on Qwen3-4B [R] https://www.reddit.com/r/MachineLearning/comments/1uv5l75/evaluating_jspace_entropy_as_an_error_predictor/
1970 reddit SHIPPING 0.280 Compressed Version of Qwen-3.6-27B coming from PrismML - Khosla-Backed Startup Claims Breakthrough With Largest-Ever AI Model on an iPhone https://www.reddit.com/r/LocalLLaMA/comments/1uv54fv/compressed_version_of_qwen3627b_coming_from/
2138 rss MODEL RELEASE 0.090 Anthropic extends free Fable 5 access for subscribers as OpenAI's GPT-5.6 Sol heats up the pricing war https://the-decoder.com/anthropic-extends-free-fable-5-access-for-subscribers-as-openais-gpt-5-6-sol-heats-up-the-pricing-war/
2319 reddit UNCATEGORIZED 0.010 The print success rates nobody talks about :Meshy vs Hi3D after 50+ models. https://www.reddit.com/r/artificial/comments/1uv50ty/the_print_success_rates_nobody_talks_about_meshy/
1979 reddit LOCAL AI 0.300 Experiment: autonomous NPCs powered by Gemma 4 E2B in the browser https://www.reddit.com/r/LocalLLaMA/comments/1uv3wnt/experiment_autonomous_npcs_powered_by_gemma_4_e2b/
2442 reddit PROBLEM SOLVED 0.260 Prompt-engineering paper accepted to ICML [R] https://www.reddit.com/r/MachineLearning/comments/1uv1xb3/promptengineering_paper_accepted_to_icml_r/
2313 reddit UNCATEGORIZED 0.010 "Is the ""J-Space"" an emergent feature, or a strategic response to optimization pressure?" https://www.reddit.com/r/artificial/comments/1uuz89v/is_the_jspace_an_emergent_feature_or_a_strategic/
2083 hackernews UNCATEGORIZED 0.020 Ask HN: Add flag for AI-generated articles https://news.ycombinator.com/item/48886741
2323 reddit INFRASTRUCTURE 0.050 AI agents may need an identity before they need more intelligence https://www.reddit.com/r/artificial/comments/1uuxhe6/ai_agents_may_need_an_identity_before_they_need/
1974 reddit PROBLEM SOLVED 0.180 Running Qwen3.5-122B on Mac Studio 96GB: Fixed 3 bugs that made long-context inference usable https://www.reddit.com/r/LocalLLaMA/comments/1uuwrc0/running_qwen35122b_on_mac_studio_96gb_fixed_3/
2151 reddit INFRASTRUCTURE 0.050 Someone built an AI agent that hacks networks and holds data for ransom. It just worked. https://www.reddit.com/r/artificial/comments/1uuouu7/someone_built_an_ai_agent_that_hacks_networks_and/
2097 hackernews UNCATEGORIZED 0.050 The One-Step Trap (In AI Research) http://incompleteideas.net/IncIdeas/OneStepTrap.html
2086 hackernews BUSINESS 0.000 I love LLMs, I hate hype https://geohot.github.io//blog/jekyll/update/2026/07/12/i-love-llms.html
2225 hackernews INFRASTRUCTURE 0.570 Show HN: Juggler – an open-source GUI coding agent, by the creator of JUCE https://github.com/juggler-ai/juggler
2095 hackernews INFRASTRUCTURE 0.020 Mechanistic interpretability researchers applying causality theory to LLMs https://cacm.acm.org/news/can-we-understand-how-large-language-models-reason/
2445 reddit PROBLEM SOLVED 0.260 Ph.D. in Operations Research / Big Tech Eng: How to transition into intermediate/advanced ML for high-value industries (Robotics, Defense, Finance)? [D] https://www.reddit.com/r/MachineLearning/comments/1uumkkg/phd_in_operations_research_big_tech_eng_how_to/
2089 hackernews PROBLEM SOLVED 0.410 Migrating a production AI agent to GPT-5.6: 2.2x faster, 27% cheaper https://ploy.ai/blog/migrating-a-production-ai-agent-to-gpt-5-6
2320 reddit UNCATEGORIZED 0.010 this openai court story is starting to look ugly https://www.reddit.com/r/artificial/comments/1uul5ef/this_openai_court_story_is_starting_to_look_ugly/
2039 rss UNCATEGORIZED 0.000 LinkedIn is the undisputed king of long-form AI slop, according to a study spanning five platforms https://the-decoder.com/linkedin-is-the-undisputed-king-of-long-form-ai-slop-according-to-a-study-spanning-five-platforms/
2037 rss MODEL RELEASE 0.090 Claude Code now has a built-in browser that lets the AI read, click, and type on external websites https://the-decoder.com/claude-code-now-has-a-built-in-browser-that-lets-the-ai-read-click-and-type-on-external-websites/
1571 reddit SHIPPING 0.190 Kreuzberg (local document extraction) is being renamed to Xberg - current version on LTS https://www.reddit.com/r/LocalLLaMA/comments/1uuhqlz/kreuzberg_local_document_extraction_is_being/
1980 reddit UNCATEGORIZED 0.170 Local Image to 3D (<2gb RAM, <20s, Apple Silicon, iPhone) https://www.reddit.com/r/LocalLLaMA/comments/1uuga40/local_image_to_3d_2gb_ram_20s_apple_silicon_iphone/
2092 hackernews PROBLEM SOLVED 0.190 AI boosts research careers but narrow the span of ideas explored: study https://spectrum.ieee.org/ai-science-research-flattens-discovery
1976 reddit LOCAL AI 0.440 If you use Open Code or other agenting programs you are leaving a lot of t/s if you don't actually use agents in parallel. Benchmark : RTX5090, Qwen3.6 35B loaded via LM studio with parallel tasks set to 8 https://www.reddit.com/r/LocalLLaMA/comments/1uueuks/if_you_use_open_code_or_other_agenting_programs/
2427 reddit SHIPPING 0.300 Zer0Fit: I took Google's new TabFM & TimesFM ML foundation models and made them available as an MCP server for zero-shot ML tasks (forecasts / classifications / regressions). 100% local. [P] https://www.reddit.com/r/MachineLearning/comments/1uue8cc/zer0fit_i_took_googles_new_tabfm_timesfm_ml/
1573 reddit LOCAL AI 0.260 I got Nemotron Puzzle 75B running smoothly on a 64GB M2 Max https://www.reddit.com/r/LocalLLaMA/comments/1uue46z/i_got_nemotron_puzzle_75b_running_smoothly_on_a/
1984 reddit UNCATEGORIZED 0.050 Working around Qwen3.6-27B's tool-call failures and looping https://www.reddit.com/r/LocalLLaMA/comments/1uue278/working_around_qwen3627bs_toolcall_failures_and/
1965 reddit SHIPPING 0.260 Zer0Fit: I took Google's new TabFM & TimesFM ML foundation models and made them available as an MCP server for zero-shot ML tasks (forecasts / classifications / regressions). 100% local. https://www.reddit.com/r/LocalLLaMA/comments/1uudxi8/zer0fit_i_took_googles_new_tabfm_timesfm_ml/
2041 rss PROBLEM SOLVED 0.170 "S&P Global sees OpenAI as a ""key credit risk"" for Oracle and cuts its credit rating" https://the-decoder.com/sp-global-sees-openai-as-a-key-credit-risk-for-oracle-and-cuts-its-credit-rating/
2431 reddit UNCATEGORIZED 0.050 Obtaining Irregular Learning Curves with HyberBand Tuned ANN model for Price Prediction [P] https://www.reddit.com/r/MachineLearning/comments/1uud3qj/obtaining_irregular_learning_curves_with/
2042 rss CULTURE 0.030 Meta kills Muse Image feature that let anyone generate AI photos of Instagram users without consent https://the-decoder.com/meta-kills-muse-image-feature-that-let-anyone-generate-ai-photos-of-instagram-users-without-consent/
2087 hackernews INFRASTRUCTURE 0.130 Old and new apps, via modern coding agents https://terrytao.wordpress.com/2026/07/11/old-and-new-apps-via-modern-coding-agents/
1968 reddit PROBLEM SOLVED 0.260 Benchmark - 4x 5060 Ti (64GB VRAM) (P2P) - Qwen3.6 27B (INT8 /w bf16 kv cache) @ 8 concurrency with SGLang. SGLang seems to handle higher concurrency better with this setup https://www.reddit.com/r/LocalLLaMA/comments/1uuc3pi/benchmark_4x_5060_ti_64gb_vram_p2p_qwen36_27b/
1923 rss MODEL RELEASE 0.090 Claude Cowork's biggest use case is the mundane office work nobody wants to own, Anthropic says https://the-decoder.com/claude-coworks-biggest-use-case-is-the-mundane-office-work-nobody-wants-to-own-anthropic-says/
1921 rss CULTURE 0.000 "OpenAI CEO Altman is now ""pretty sure"" AI is net job-creating, which is quite the pivot from predicting mass layoffs" https://the-decoder.com/openai-ceo-altman-is-now-pretty-sure-ai-is-net-job-creating-which-is-quite-the-pivot-from-predicting-mass-layoffs/
1977 reddit LOCAL AI 0.450 Voodoo Quant beats Unsloth Dynamic 2.0 KLD by 95% in Qwen3.5 0.8B and 2B https://www.reddit.com/r/LocalLLaMA/comments/1uua3jd/voodoo_quant_beats_unsloth_dynamic_20_kld_by_95/
1924 rss UNCATEGORIZED 0.000 Grades dropped from 96 to 48 percent when a Brown professor made students take the exam without AI https://the-decoder.com/grades-dropped-from-96-to-48-percent-when-a-brown-professor-made-students-take-the-exam-without-ai/
1824 rss INFRASTRUCTURE 0.040 AI agents win at Slay the Spire 2 after researchers replace growing chat logs with structured memory https://the-decoder.com/ai-agents-win-at-slay-the-spire-2-after-researchers-replace-growing-chat-logs-with-structured-memory/
1564 reddit MODEL RELEASE 0.100 Need help tuning cache in llama-server https://www.reddit.com/r/LocalLLaMA/comments/1uu8g9f/need_help_tuning_cache_in_llamaserver/
2094 hackernews INFRASTRUCTURE 0.180 Show HN: Mindwalk – Replay coding-agent sessions on a 3D map of your codebase https://github.com/cosmtrek/mindwalk
1569 reddit PROBLEM SOLVED 0.180 i would like to share my experience. working with huge LLMs and poor Machine https://www.reddit.com/r/LocalLLaMA/comments/1uu6qvh/i_would_like_to_share_my_experience_working_with/
1975 reddit INFRASTRUCTURE 0.320 **Your $80 Tesla P100 has been doing silently noisy math in llama.cpp for years. Three lines fix it, for free.** https://www.reddit.com/r/LocalLLaMA/comments/1uu6p9o/your_80_tesla_p100_has_been_doing_silently_noisy/
1978 reddit UNCATEGORIZED 0.010 I mapped Anthropic’s J-Space Hallucination signal across 7 datasets on Qwen3-4B to find out where it works and where it breaks https://www.reddit.com/r/LocalLLaMA/comments/1uu61wb/i_mapped_anthropics_jspace_hallucination_signal/
1566 reddit INFRASTRUCTURE 0.120 First attempts at a CPU setup - MS-02 Intel 285hx, trying Qwen3, Qwen3.6 and Gemma4 https://www.reddit.com/r/LocalLLaMA/comments/1uu5ht0/first_attempts_at_a_cpu_setup_ms02_intel_285hx/
1966 reddit UNCATEGORIZED 0.010 I didn't give up - extGemma4-40_5B returned https://www.reddit.com/r/LocalLLaMA/comments/1uu4hxp/i_didnt_give_up_extgemma440_5b_returned/
1572 reddit LOCAL AI 0.300 Qwenthropic https://www.reddit.com/r/LocalLLaMA/comments/1uu3545/qwenthropic/
1983 reddit LOCAL AI 0.300 Interactive Jacobian-Lens visualizer and live steerer for GGUF models on llama.cpp https://www.reddit.com/r/LocalLLaMA/comments/1uu32z6/interactive_jacobianlens_visualizer_and_live/
2085 hackernews MODEL RELEASE 0.270 What xAI's Grok build CLI sends to xAI: A wire-level analysis https://gist.github.com/cereblab/dc9a40bc26120f4540e4e09b75ffb547
1565 reddit LOCAL AI 0.340 Measuring PCIe transfer under dual GPU with pipeline & tensor llama.cpp https://www.reddit.com/r/LocalLLaMA/comments/1utz50z/measuring_pcie_transfer_under_dual_gpu_with/
2088 hackernews UNCATEGORIZED 0.020 Mesh LLM: distributed AI computing on iroh https://www.iroh.computer/blog/mesh-llm
2090 hackernews UNCATEGORIZED 0.020 Stop Telling Me to Ask an LLM https://blog.yaelwrites.com/stop-telling-me-to-ask-an-llm/
1967 reddit PROBLEM SOLVED 0.180 Ultra budget 20GB vram with 448GB/s for $100 bucks. https://www.reddit.com/r/LocalLLaMA/comments/1utwqf8/ultra_budget_20gb_vram_with_448gbs_for_100_bucks/
1559 reddit INFRASTRUCTURE 0.340 Performance comparison on full compute performance (Anima) and LLM prompt processing of 5090 (600,475 and 400W) vs 6000 PRO MaxQ shunt modded and water cooled (at 300, 400, 475 and 600W), and 6000 PRO WS/SE (600W). https://www.reddit.com/r/LocalLLaMA/comments/1utvbey/performance_comparison_on_full_compute/
1558 reddit LOCAL AI 0.500 I benched quad 5060Tis for code generation with Qwen3.6-27B so you don't have to (it's really good) https://www.reddit.com/r/LocalLLaMA/comments/1uturng/i_benched_quad_5060tis_for_code_generation_with/
2096 hackernews BUSINESS 0.020 Wealthy AI workers send San Francisco house prices soaring https://www.bbc.com/news/articles/c9q29j47v9ro
1992 hackernews CULTURE 0.020 AI 2040 and the cult of intelligence https://geohot.github.io//blog/jekyll/update/2026/07/11/ai-2040.html
2000 hackernews INFRASTRUCTURE 0.100 Who manages the agents? https://www.off-policy.com/dont-go-quietly-into-the-ai-night/
1828 rss UNCATEGORIZED 0.000 OpenAI's GPT-5.6 Sol Ultra reportedly solves a 50-year-old math problem in under an hour https://the-decoder.com/openais-gpt-5-6-sol-ultra-reportedly-solves-a-50-year-old-math-problem-in-under-an-hour/
1999 hackernews CULTURE 0.020 Reverse centaurs are the answer to the AI paradox (2025) https://pluralistic.net/2025/09/11/vulgar-thatcherism/#there-is-an-alternative
1731 rss MODEL RELEASE 0.090 Terrorist groups are using every major AI chatbot for attack planning and weapons development https://the-decoder.com/terrorist-groups-are-using-every-major-ai-chatbot-for-attack-planning-and-weapons-development/
2001 hackernews INFRASTRUCTURE 0.180 Show HN: Reame – a CPU inference server that gets faster as it runs https://github.com/swellweb/reame
1730 rss BUSINESS 0.000 OpenAI bets on families as ChatGPT goes deeper into households https://techcrunch.com/2026/07/11/openai-bets-on-families-as-chatgpt-goes-deeper-into-households/
1776 hackernews UNCATEGORIZED 0.020 Ghost Font: A font that humans can read but AI cannot https://www.mixfont.com/ghost-font
1786 hackernews BUSINESS 0.020 Microsoft latest report shows 25% emissions raised due to AI data centers https://www.windowscentral.com/microsoft/dropping-greenwashing-credits-and-expanding-ai-datacenters-caused-microsofts-25-percent-emissions-jump
1592 hackernews PROBLEM SOLVED 0.190 Companies are scrambling to curtail soaring AI costs https://www.economist.com/business/2026/06/14/companies-are-scrambling-to-curtail-soaring-ai-costs
1591 hackernews SHIPPING 0.230 Meta pulls new AI image feature after days of backlash https://www.bbc.com/news/articles/c2dy6e8klw0o
1590 hackernews PROBLEM SOLVED 0.190 AI Can't Recreate the Thrust Game (But It Can Help You Understand It) https://www.jamesdrandall.com/posts/thrust_ai_powered_software_archaeology/
1588 hackernews CULTURE 0.020 Apple sues OpenAI, accusing it of stealing company secrets https://www.nytimes.com/2026/07/10/technology/apple-openai-lawsuit.html
1 id source bucket final_score verdict title url
2 2702 rss SHIPPING 0.180 OpenAI researcher Miles Wang in talks to launch AI drug discovery startup valued at $2B https://techcrunch.com/2026/07/14/openai-researcher-miles-wang-in-talks-to-launch-ai-drug-discovery-startup-valued-at-2b/
3 2587 rss CULTURE 0.000 Lorde says AI glasses are ‘not sexy’ https://techcrunch.com/2026/07/14/lorde-says-ai-glasses-are-not-sexy/
4 2584 rss UNCATEGORIZED 0.030 OpenAI’s first hardware device is reportedly a screenless speaker that can move https://techcrunch.com/2026/07/14/openais-first-hardware-device-is-reportedly-a-screenless-speaker-that-can-move/
5 2589 rss CULTURE 0.000 OpenAI pushes back on Apple trade secret lawsuit https://techcrunch.com/2026/07/14/openai-pushes-back-on-apple-trade-secret-lawsuit/
6 2672 hackernews UNCATEGORIZED 0.020 Financing the AI boom: from cash flows to debt [pdf] https://www.bis.org/publ/bisbull120.pdf
7 2585 rss MODEL RELEASE 0.120 OpenAI’s new flagship model deletes files on its own, people keep warning https://techcrunch.com/2026/07/14/openais-new-flagship-model-deletes-files-on-its-own-people-keep-warning/
8 2646 reddit MODEL RELEASE 0.130 Opening the Black Box: Unison Zero Parameter Model https://www.reddit.com/r/artificial/comments/1uwjwl6/opening_the_black_box_unison_zero_parameter_model/
9 2648 reddit PROBLEM SOLVED 0.220 Developers Hate AI. I Used It To Sell 10 Websites This Week. https://www.reddit.com/r/artificial/comments/1uwj75g/developers_hate_ai_i_used_it_to_sell_10_websites/
10 2485 rss SHIPPING 0.210 Apple opens its new Siri AI to everyone with the iOS 27 public beta https://techcrunch.com/2026/07/14/apple-opens-its-new-siri-ai-to-everyone-with-the-ios-27-public-beta/
11 2486 rss CULTURE 0.000 Anthropic’s newest ad is creeping people out https://techcrunch.com/2026/07/14/anthropics-newest-ad-is-creeping-people-out/
12 2487 rss BUSINESS 0.030 The founder of Hinge raised $18M to build a new AI dating service, Overtone https://techcrunch.com/2026/07/14/the-founder-of-hinge-raised-18m-to-build-a-new-ai-dating-service-overtone/
13 2654 reddit BUSINESS 0.010 How does a 102M-parameter transformer forecast multivariate time series? https://www.reddit.com/r/artificial/comments/1uwh9ko/how_does_a_102mparameter_transformer_forecast/
14 2490 rss MODEL RELEASE 0.090 Google faces another AI training lawsuit from major publishers https://techcrunch.com/2026/07/14/google-faces-another-ai-training-lawsuit-from-major-publishers/
15 2652 reddit CULTURE 0.000 Apple just sued OpenAI for trade secret theft. And Google quietly rewrote how the internet works. https://www.reddit.com/r/artificial/comments/1uwh06x/apple_just_sued_openai_for_trade_secret_theft_and/
16 2645 reddit SHIPPING 0.230 The absolute nightmare of putting AI agents into actual production https://www.reddit.com/r/artificial/comments/1uwg8kk/the_absolute_nightmare_of_putting_ai_agents_into/
17 2624 arxiv RESEARCH 0.190 Do AI Agents Know When a Task Is Simple? Toward Complexity-Aware Reasoning and Execution https://arxiv.org/abs/2607.13034v1
18 2625 arxiv RESEARCH 0.150 The Seriality Gap in Video Diffusion Models https://arxiv.org/abs/2607.13031v1
19 2626 arxiv RESEARCH 0.340 TerraZero: Procedural Driving Simulation for Zero-Demonstration Self-Play at Scale https://arxiv.org/abs/2607.13028v1
20 2627 arxiv RESEARCH 0.340 PalmClaw: A Native On-Device Agent Framework for Mobile Phones https://arxiv.org/abs/2607.13027v1
21 2628 arxiv RESEARCH 0.150 A Shortcut to Statistically Steady-State Turbulence with Flow Matching https://arxiv.org/abs/2607.13022v1
22 2649 reddit PROBLEM SOLVED 0.210 Ford replaced engineers with AI, then quietly hired 350 back. The reason should stop every founder about to cut their team to SAVE money. https://www.reddit.com/r/artificial/comments/1uwg31g/ford_replaced_engineers_with_ai_then_quietly/
23 2629 arxiv RESEARCH 0.110 Audio-Native Speech Recognition with a Frozen Discrete-Diffusion Language Model https://arxiv.org/abs/2607.13013v1
24 2630 arxiv RESEARCH 0.140 Dynamic Resource Allocation for Ensemble Determinization MCTS https://arxiv.org/abs/2607.13007v1
25 2631 arxiv RESEARCH 0.180 The Spectrum Is Not Enough: When Context Helps Time-Series Forecasting https://arxiv.org/abs/2607.13006v1
26 2632 arxiv RESEARCH 0.150 Watermark Forensics for Generative Models: An Information-Theoretic Perspective https://arxiv.org/abs/2607.13003v1
27 2494 rss MODEL RELEASE 0.090 DeepMind CEO calls for an independent standards body to regulate frontier AI https://techcrunch.com/2026/07/14/deepmind-ceo-calls-for-an-independent-standards-body-to-regulate-frontier-ai/
28 2429 reddit MODEL RELEASE 0.210 [P] RL-training Qwen3.6 to RL-train tool using AI models [P] https://www.reddit.com/r/MachineLearning/comments/1uwfmfa/p_rltraining_qwen36_to_rltrain_tool_using_ai/
29 2633 arxiv RESEARCH 0.110 Win by Silence: Deletion Non-Monotonicity, Autonomous Exploitation, and Typed-State Gating in LLM Plan Evaluation https://arxiv.org/abs/2607.12986v1
30 2634 arxiv RESEARCH 0.110 Resist and Update: Counterfactual Report Coordinates for Incentive-Compatible LLMs https://arxiv.org/abs/2607.12985v1
31 2635 arxiv RESEARCH 0.140 FormalAnalyticGeo: A Neural-Symbolic Based Framework for Multimodal Analytic Geometry Problem Generation https://arxiv.org/abs/2607.12982v1
32 2488 rss MODEL RELEASE 0.090 Anthropic opens Claude for Teachers with a promise not to train models on student data https://the-decoder.com/anthropic-opens-claude-for-teachers-with-a-promise-not-to-train-models-on-student-data/
33 2636 arxiv RESEARCH 0.180 Ensemble Controlled-Flow Filtering for Implicit Data Assimilation https://arxiv.org/abs/2607.12975v1
34 2637 arxiv RESEARCH 0.210 The Illusion of Robustness: Aggregate Accuracy Hides Prediction Flips under Task-Irrelevant Context https://arxiv.org/abs/2607.12963v1
35 2638 arxiv RESEARCH 0.440 Form, Not Content? A Preregistered, Placebo-Controlled Evaluation of Learned Error-Conditioned Self-Repair Through Prompts and Weights in Frozen Small Code Models https://arxiv.org/abs/2607.12962v1
36 2656 reddit SHIPPING 0.190 Structured output reliability with LLMs — 3-month production learnings https://www.reddit.com/r/artificial/comments/1uwe9qp/structured_output_reliability_with_llms_3month/
37 2639 arxiv RESEARCH 0.190 Robustness of Deep Learning Models for PV Power Forecasting under NWP Forecast Errors: A Spatiotemporal and Physically Interpretable Analysis https://arxiv.org/abs/2607.12954v1
38 2640 arxiv RESEARCH 0.250 ViHoRec: A Quality-Controlled Vietnamese Hotel Recommendation Dataset and Cold-Start Benchmark https://arxiv.org/abs/2607.12946v1
39 2502 rss BUSINESS 0.030 DeepSeek needs more cash just weeks after closing its first $7 billion round https://the-decoder.com/deepseek-needs-more-cash-just-weeks-after-closing-its-first-7-billion-round/
40 2168 rss CULTURE 0.000 Meta’s Adam Mosseri says AI token budgets could soon be capped per engineer https://techcrunch.com/2026/07/14/metas-adam-mosseri-says-ai-token-budgets-could-soon-be-capped-per-engineer/
41 2164 rss UNCATEGORIZED 0.030 Google Search now generates AI images when it can't find what you're looking for on the web https://the-decoder.com/google-search-now-generates-ai-images-when-it-cant-find-what-youre-looking-for-on-the-web/
42 2150 reddit MODEL RELEASE 0.140 All cross thread implementation of memory in chatgpt, claude, and gemini is unsafe https://www.reddit.com/r/artificial/comments/1uwdc0k/all_cross_thread_implementation_of_memory_in/
43 2641 arxiv RESEARCH 0.180 Efficient Sequential Calibration with $O(T^{2/3-ε})$ Error Bound https://arxiv.org/abs/2607.12928v1
44 2166 rss UNCATEGORIZED 0.000 Google Images gets a Pinterest-like redesign focused on discovery https://techcrunch.com/2026/07/14/google-images-gets-a-pinterest-like-redesign-focused-on-discovery/
45 2642 arxiv RESEARCH 0.110 Knowledge- and Gradient-Guided Reinforcement Learning for Parametrized Action Markov Decision Processes https://arxiv.org/abs/2607.12924v1
46 2643 arxiv RESEARCH 0.180 LatentFlow: A General Framework for Conditioning Stochastic Processes https://arxiv.org/abs/2607.12922v1
47 2167 rss UNCATEGORIZED 0.000 AWS and Bluesight build AI for hospital 340B compliance https://www.artificialintelligence-news.com/news/aws-and-bluesight-build-ai-for-hospital-340b-compliance/
48 2315 reddit SHIPPING 0.460 Open Source Local LLM Training Tool (for consumer hardware) https://www.reddit.com/r/artificial/comments/1uwcah2/open_source_local_llm_training_tool_for_consumer/
49 2441 reddit PROBLEM SOLVED 0.330 New LLM Coordination Benchmark - Benchmarking Open-Ended Multi-Agent Coordination in Language Agents [R] https://www.reddit.com/r/MachineLearning/comments/1uwc6ni/new_llm_coordination_benchmark_benchmarking/
50 2326 reddit INFRASTRUCTURE 0.050 I'm not a great artist — so I made an agent that turns my doodles on my Remarkable tablet into actually nice charcoal sketches. Real editable pen-line vectors too! Not just static images. https://www.reddit.com/r/artificial/comments/1uwbt7o/im_not_a_great_artist_so_i_made_an_agent_that/
51 2219 hackernews UNCATEGORIZED 0.020 Are we offloading too much of our thinking to AI? https://www.artfish.ai/p/offloading-thinking-to-ai
52 2257 rss PROBLEM SOLVED 0.200 New York State halts construction of all new data centers https://techcrunch.com/2026/07/14/new-york-state-halts-construction-of-all-new-data-centers/
53 2325 reddit INFRASTRUCTURE 0.010 A new, state-of-the-art, agentic pipeline for easy Music Video creation https://www.reddit.com/r/artificial/comments/1uwbfos/a_new_stateoftheart_agentic_pipeline_for_easy/
54 2562 hackernews UNCATEGORIZED 0.020 The Agentic Loop: Three loops in a trench coat https://www.bobbytables.io/p/the-agentic-loop-three-loops-in-a
55 2253 rss BUSINESS 0.290 Reflection inks $1B compute deal with Nebius https://techcrunch.com/2026/07/14/reflection-inks-1b-compute-deal-with-nebius/
56 2254 rss CULTURE 0.290 The real AI race may no longer be at the frontier https://techcrunch.com/2026/07/14/the-real-ai-race-may-no-longer-be-at-the-frontier-open-models-hugging-face/
57 2255 rss UNCATEGORIZED 0.030 Spotify expands its AI push with a ChatGPT-like music assistant https://techcrunch.com/2026/07/14/spotify-expands-its-ai-push-with-a-chatgpt-like-music-assistant/
58 2256 rss UNCATEGORIZED 0.030 Superhuman’s new auto-draft feature almost makes me like AI replies https://techcrunch.com/2026/07/14/superhumans-new-auto-draft-feature-almost-makes-me-like-ai-replies/
59 2314 reddit INFRASTRUCTURE 0.190 The real bottleneck for AI agents may be proving who they are https://www.reddit.com/r/artificial/comments/1uw81un/the_real_bottleneck_for_ai_agents_may_be_proving/
60 2217 hackernews UNCATEGORIZED 0.020 Proof of care in the age of AI https://jacobfilipp.com/care/
61 2461 hackernews MODEL RELEASE 0.270 Show HN: I RL-trained an agent that trains models with RL (for ~$1.3k) https://github.com/Danau5tin/ai-trains-ai
62 2224 hackernews INFRASTRUCTURE 0.060 Coding agents think ahead of time https://arxiv.org/abs/2607.05188
63 2324 reddit BUSINESS 0.010 Did you know the CEO of OpenAI owns nearly 9% of Reddit while Reddit bans users for AI generated content? https://www.reddit.com/r/artificial/comments/1uw6sv6/did_you_know_the_ceo_of_openai_owns_nearly_9_of/
64 2258 rss CULTURE 0.000 ChatGPT returns to WhatsApp in Europe after EU forces Meta to open the door to rival AI bots https://the-decoder.com/chatgpt-returns-to-whatsapp-in-europe-after-eu-forces-meta-to-open-the-door-to-rival-ai-bots/
65 2260 rss PROBLEM SOLVED 0.240 Deepmind CEO Hassabis says "nobody in the world knows what happens next" so "cautious optimism" means building guardrails now https://the-decoder.com/deepmind-ceo-hassabis-says-nobody-in-the-world-knows-what-happens-next-so-cautious-optimism-means-building-guardrails-now/
66 2157 hackernews UNCATEGORIZED 0.140 Codex starts encrypting sub-agent prompts https://github.com/openai/codex/issues/28058
67 2262 rss BUSINESS 0.000 PixVerse's $2B valuation shows investors still believe AI video generation has room for another winner https://the-decoder.com/pixverses-2b-valuation-shows-investors-still-believe-ai-video-generation-has-room-for-another-winner/
68 2264 rss MODEL RELEASE 0.120 Claude responds with more warmth in Hindi and more rigor in Russian, showing how language shapes AI answers https://the-decoder.com/claude-values-study/
69 2220 hackernews UNCATEGORIZED 0.020 Demis Hassabis has a plan to harness AI safely https://twitter.com/demishassabis/status/2076957440109625718
70 2317 reddit UNCATEGORIZED 0.040 The first AI was a syllogism machine in 1956. We're still building the same thing. https://www.reddit.com/r/artificial/comments/1uw23qw/the_first_ai_was_a_syllogism_machine_in_1956_were/
71 2460 hackernews BUSINESS 0.020 OpenAI's Ad Business Is on Pace to Miss Its Own Forecast by 90%, Analyst Says https://www.adweek.com/media/openais-ad-business-is-on-pace-to-miss-its-own-forecast-by-90-analyst-says/
72 2432 reddit UNCATEGORIZED 0.050 How many on-the-fly augmentations per image for a single-class segmentation mode [R] https://www.reddit.com/r/MachineLearning/comments/1uvxt70/how_many_onthefly_augmentations_per_image_for_a/
73 2327 reddit UNCATEGORIZED 0.040 Inside Ghostcommit: How Malicious PNGs Bypass AI Code Reviewers https://www.reddit.com/r/artificial/comments/1uvxqg5/inside_ghostcommit_how_malicious_pngs_bypass_ai/
74 2152 reddit INFRASTRUCTURE 0.000 We keep asking whether AI will replace us. The more useful question is what it means to share the world with it. https://www.reddit.com/r/artificial/comments/1uvvd13/we_keep_asking_whether_ai_will_replace_us_the/
75 2266 rss UNCATEGORIZED 0.030 Uber’s product chief on hotels, robotaxis, and why the company doesn’t want to be ‘everything for everyone’ https://techcrunch.com/2026/07/13/ubers-product-chief-on-hotels-robotaxis-and-why-the-company-doesnt-want-to-be-everything-for-everyone/
76 2267 rss BUSINESS 0.000 Video-generation startup PixVerse raises $439M, valuation soars past $2B https://techcrunch.com/2026/07/13/video-generation-startup-pixverse-raises-439m-valuation-soars-past-2b/
77 2322 reddit MODEL RELEASE 0.100 Anthropic analyzed 300,000 real Claude conversations to measure its values. The findings are uncomfortable. https://www.reddit.com/r/artificial/comments/1uvpob7/anthropic_analyzed_300000_real_claude/
78 2158 hackernews UNCATEGORIZED 0.020 Samsung Health app threatens data deletion if users opt out AI training https://neow.in/cWsyMTV3
79 2226 hackernews UNCATEGORIZED 0.140 Show HN: I implemented a neural network in SQL https://github.com/xqlsystems/xarray-sql/blob/claude/xarray-sql-mnist-demo/benchmarks/nn.py
80 2221 hackernews UNCATEGORIZED 0.020 AI is a bad tool https://bytecode.news/posts/2026/07/user-submission-ai-is-a-bad-tool
81 2438 reddit PROBLEM SOLVED 0.410 GPUHedge: Hedging serverless GPU providers improves cold start p95 latency from 117s to 30s [P] https://www.reddit.com/r/MachineLearning/comments/1uvlb6h/gpuhedge_hedging_serverless_gpu_providers/
82 2223 hackernews INFRASTRUCTURE 0.060 Show HN: Nobie – an Excel-compatible runtime for agents and humans https://nobie.com
83 2126 rss CULTURE 0.000 The wildest allegations in Apple’s trade secrets lawsuit against OpenAI https://techcrunch.com/2026/07/13/the-wildest-allegations-in-apples-trade-secrets-lawsuit-against-openai/
84 2463 hackernews INFRASTRUCTURE 0.180 Show HN: BillAI Bass, an AI-Powered Big Mouth Billy Bass Using Strands Agents https://github.com/morganwilliscloud/billai-bass
85 2098 hackernews MODEL RELEASE 0.150 xAI's Grok Build CLI Uploads Git Repositories to a Google Cloud Bucket https://www.internationalcyberdigest.com/xais-grok-build-cli-uploads-entire-git-repositories-to-a-google-cloud-bucket/
86 2125 rss UNCATEGORIZED 0.000 What Anthropic’s latest AI discovery does—and doesn’t—show https://www.technologyreview.com/2026/07/13/1140343/what-anthropics-latest-ai-discovery-does-and-doesnt-show/
87 2144 arxiv RESEARCH 0.180 Requential Coding: Pushing the Limits of Model Compression with Self-Generated Training Data https://arxiv.org/abs/2607.11883v1
88 2145 arxiv RESEARCH 0.110 Metacognition in LLMs: Foundations, Progress, and Opportunities https://arxiv.org/abs/2607.11881v1
89 2146 arxiv RESEARCH 0.180 Invariant Learning Dynamics of Transformers in Inductive Reasoning Tasks https://arxiv.org/abs/2607.11875v1
90 2147 arxiv RESEARCH 0.300 A Minimalist Retargeting-Guided Reinforcement Learning Recipe for Dexterous Manipulation https://arxiv.org/abs/2607.11874v1
91 2148 arxiv RESEARCH 0.290 A Durability and Cross-Language Transfer Benchmark for a Validated Teaching-Feedback Classification Protocol https://arxiv.org/abs/2607.11873v1
92 2194 arxiv RESEARCH 0.150 Inside the Unfair Judge: A Mechanistic Interpretability Account of LLM-as-Judge Bias https://arxiv.org/abs/2607.11871v1
93 2444 reddit CULTURE 0.050 Chain of Thought is a scaling trap. the next wave is latent reasoning (Coconut / HRM / RecrusiveMAS)... but then we hit the black box wall. Where does BDH fit? [D] https://www.reddit.com/r/MachineLearning/comments/1uviru5/chain_of_thought_is_a_scaling_trap_the_next_wave/
94 2316 reddit INFRASTRUCTURE 0.050 The 'agent web' is coming — where AI agents talk directly to each other instead of scraping websites https://www.reddit.com/r/artificial/comments/1uviqvw/the_agent_web_is_coming_where_ai_agents_talk/
95 2195 arxiv RESEARCH 0.280 Evidence-Backed Video Question Answering https://arxiv.org/abs/2607.11862v1
96 2196 arxiv RESEARCH 0.280 AdvancedMathBench: A Benchmark Suite for Advanced Mathematical Proof Generation and Verification https://arxiv.org/abs/2607.11849v1
97 2197 arxiv RESEARCH 0.180 Input-Aware Dynamic Backdoor Attack Against Quantum Neural Networks https://arxiv.org/abs/2607.11843v1
98 2121 rss BUSINESS 0.000 Sam Altman’s space data center trash talk is what most experts already believe https://techcrunch.com/2026/07/13/sam-altmans-space-data-center-trash-talk-is-what-most-experts-already-believe/
99 2198 arxiv RESEARCH 0.110 LoRA-Based Cascaded Multimodal Fusion for Action Recognition in Medical Training Environments https://arxiv.org/abs/2607.11839v1
100 2199 arxiv RESEARCH 0.300 Transformer-Guided Swarm Intelligence for Frugal Neural Architecture Search https://arxiv.org/abs/2607.11826v1
101 2123 rss INFRASTRUCTURE 0.110 Turing Award winner Rich Sutton founds Oak Lab to build AI agents that learn on their own https://the-decoder.com/turing-award-winner-rich-sutton-founds-oak-lab-to-build-ai-agents-that-learn-on-their-own/
102 2200 arxiv RESEARCH 0.360 MM-ToolSandBox: A Unified Framework for Evaluating Visual Tool-Calling Agents https://arxiv.org/abs/2607.11818v1
103 2201 arxiv RESEARCH 0.150 Relaxing Faithfulness with Intervention-Only Causal Discovery https://arxiv.org/abs/2607.11816v1
104 2202 arxiv RESEARCH 0.110 Introducing Human-Centeredness in AI-Assisted Lexicography https://arxiv.org/abs/2607.11808v1
105 2203 arxiv RESEARCH 0.110 Encoder-Side Neuron Identification and Amplification for Acoustic Perception in Large Audio-Language Models https://arxiv.org/abs/2607.11801v1
106 2204 arxiv RESEARCH 0.140 StoryTeller: Training-Free Narrative Grounding for Long-Form Audio Description https://arxiv.org/abs/2607.11798v1
107 2205 arxiv RESEARCH 0.150 An Exact Instrument for State Usage in Selective State-Space Models, and the Input-Driven Migration It Reveals https://arxiv.org/abs/2607.11796v1
108 2318 reddit UNCATEGORIZED 0.010 Is there any kind of AI that could "read" huge loads of emails and give a "mark" according to a given expected result? https://www.reddit.com/r/artificial/comments/1uvgqrn/is_there_any_kind_of_ai_that_could_read_huge/
109 2206 arxiv RESEARCH 0.150 Forgetting Our Way to Shared Meaning: Effects of Forgetting on Conceptual Alignment in a Non-Partnership Coordination Game https://arxiv.org/abs/2607.11787v1
110 2207 arxiv RESEARCH 0.210 How Temperature Shapes Ideological Discourse in Retrieval-Augmented Generation? https://arxiv.org/abs/2607.11783v1
111 2130 rss UNCATEGORIZED 0.000 Should AI help you get away with killing your spouse? https://techcrunch.com/2026/07/13/should-ai-help-you-get-away-with-killing-your-spouse/
112 2208 arxiv RESEARCH 0.110 Evaluating RE Practices for Explainability: Synthesizing Insights from Daimler Truck into an Explainable RE Framework Proposal https://arxiv.org/abs/2607.11771v1
113 2129 rss UNCATEGORIZED 0.000 Nobel laureates and AI leaders warn the window to prepare for AI's economic impact is closing fast https://the-decoder.com/nobel-laureates-and-ai-leaders-warn-the-window-to-prepare-for-ais-economic-impact-is-closing-fast/
114 2222 hackernews UNCATEGORIZED 0.140 Show HN: Jacquard, a programming language for AI-written, human-reviewed code https://github.com/jbwinters/jacquard-lang
115 2131 rss MODEL RELEASE 0.090 Anthropic starts localizing Claude pricing for India, its biggest market after the US https://techcrunch.com/2026/07/13/anthropic-starts-localizing-claude-pricing-for-india-its-biggest-market-after-the-us/
116 1982 reddit LOCAL AI 0.330 Upgrade path for ryzen 9 (64 gb) + rtx 5080 https://www.reddit.com/r/LocalLLaMA/comments/1uvelii/upgrade_path_for_ryzen_9_64_gb_rtx_5080/
117 1972 reddit UNCATEGORIZED 0.010 Wan-Dancer: A Hierarchical Framework for Minute-scale Coherent Music-to-Dance Generation https://www.reddit.com/r/LocalLLaMA/comments/1uvdaq7/wandancer_a_hierarchical_framework_for/
118 2136 rss MODEL RELEASE 0.090 Nadella calls out AI labs like OpenAI and Anthropic for banning distillation while training on everyone else's data https://the-decoder.com/nadella-calls-out-ai-labs-like-openai-and-anthropic-for-banning-distillation-while-training-on-everyone-elses-data/
119 2127 rss MODEL RELEASE 0.120 Waze adds new AI-powered features and customization updates https://techcrunch.com/2026/07/13/waze-adds-new-ai-powered-features-and-customization-updates/
120 1969 reddit INFRASTRUCTURE 0.050 I benchmarked 15 "E-Waste" GPUs with Modern Workloads https://www.reddit.com/r/LocalLLaMA/comments/1uvcjd0/i_benchmarked_15_ewaste_gpus_with_modern_workloads/
121 2091 hackernews INFRASTRUCTURE 0.180 Show HN: Clawk – Give coding agents a disposable Linux VM, not your laptop https://github.com/clawkwork/clawk
122 2430 reddit SHIPPING 0.420 Hundreds of papers hit arXiv every day and maybe 3 matter to my research, so I built an open-source tool that finds them [P] https://www.reddit.com/r/MachineLearning/comments/1uvcdf7/hundreds_of_papers_hit_arxiv_every_day_and_maybe/
123 2093 hackernews MODEL RELEASE 0.110 Grok uploaded my user directory to xAI's servers https://twitter.com/a_green_being/status/2076598897779020159
124 1981 reddit UNCATEGORIZED 0.050 MCP…. Is bad? https://www.reddit.com/r/LocalLLaMA/comments/1uvaqxp/mcp_is_bad/
125 1973 reddit SHIPPING 0.350 Production Qwen 3.6-27B VLLM config? https://www.reddit.com/r/LocalLLaMA/comments/1uvacno/production_qwen_3627b_vllm_config/
126 2153 reddit PROBLEM SOLVED 0.120 I built a full 3D open-world racing game almost entirely with AI, and it now has real daily players. Here's the honest breakdown of what the model nailed and where it completely fell apart. https://www.reddit.com/r/artificial/comments/1uvaaf4/i_built_a_full_3d_openworld_racing_game_almost/
127 1971 reddit MODEL RELEASE 0.100 [Study/Models] Flint: Compressing Reasoning Without Breaking It https://www.reddit.com/r/LocalLLaMA/comments/1uv9o2u/studymodels_flint_compressing_reasoning_without/
128 2321 reddit CULTURE 0.000 Everyone keeps asking if AI will replace people. I think we’re asking the wrong question. https://www.reddit.com/r/artificial/comments/1uv9l8w/everyone_keeps_asking_if_ai_will_replace_people_i/
129 2133 rss MODEL RELEASE 0.190 German AI consortium releases Soofi S, an open 30B model that tops benchmarks in both English and German https://the-decoder.com/german-ai-consortium-releases-soofi-s-an-open-30b-model-that-tops-benchmarks-in-both-english-and-german/
130 2134 rss PROBLEM SOLVED 0.250 AI agent crawlers now need permission. Here’s how to get it https://www.artificialintelligence-news.com/news/ai-agent-crawlers-cloudflare-rules/
131 2035 rss MODEL RELEASE 0.090 Google’s SensorFM turns messy wearable sensor data into a general-purpose health intelligence layer https://the-decoder.com/sensorfm/
132 2149 reddit UNCATEGORIZED 0.040 For a silent revolution in the singularity scene https://www.reddit.com/r/artificial/comments/1uv63ms/for_a_silent_revolution_in_the_singularity_scene/
133 2084 hackernews UNCATEGORIZED 0.020 Zig Creator Calls Spade a Spade, Anthropic Blows Smoke https://raymyers.org/post/zed-creator-calls-spade-a-spade/
134 2433 reddit UNCATEGORIZED 0.050 Evaluating J-space entropy as an error predictor across 7 datasets on Qwen3-4B [R] https://www.reddit.com/r/MachineLearning/comments/1uv5l75/evaluating_jspace_entropy_as_an_error_predictor/
135 1970 reddit SHIPPING 0.280 Compressed Version of Qwen-3.6-27B coming from PrismML - Khosla-Backed Startup Claims Breakthrough With Largest-Ever AI Model on an iPhone https://www.reddit.com/r/LocalLLaMA/comments/1uv54fv/compressed_version_of_qwen3627b_coming_from/
136 2138 rss MODEL RELEASE 0.090 Anthropic extends free Fable 5 access for subscribers as OpenAI's GPT-5.6 Sol heats up the pricing war https://the-decoder.com/anthropic-extends-free-fable-5-access-for-subscribers-as-openais-gpt-5-6-sol-heats-up-the-pricing-war/
137 2319 reddit UNCATEGORIZED 0.010 The print success rates nobody talks about :Meshy vs Hi3D after 50+ models. https://www.reddit.com/r/artificial/comments/1uv50ty/the_print_success_rates_nobody_talks_about_meshy/
138 1979 reddit LOCAL AI 0.300 Experiment: autonomous NPCs powered by Gemma 4 E2B in the browser https://www.reddit.com/r/LocalLLaMA/comments/1uv3wnt/experiment_autonomous_npcs_powered_by_gemma_4_e2b/
139 2442 reddit PROBLEM SOLVED 0.260 Prompt-engineering paper accepted to ICML [R] https://www.reddit.com/r/MachineLearning/comments/1uv1xb3/promptengineering_paper_accepted_to_icml_r/
140 2313 reddit UNCATEGORIZED 0.010 Is the "J-Space" an emergent feature, or a strategic response to optimization pressure? https://www.reddit.com/r/artificial/comments/1uuz89v/is_the_jspace_an_emergent_feature_or_a_strategic/
141 2083 hackernews UNCATEGORIZED 0.020 Ask HN: Add flag for AI-generated articles https://news.ycombinator.com/item/48886741
142 2323 reddit INFRASTRUCTURE 0.050 AI agents may need an identity before they need more intelligence https://www.reddit.com/r/artificial/comments/1uuxhe6/ai_agents_may_need_an_identity_before_they_need/
143 1974 reddit PROBLEM SOLVED 0.180 Running Qwen3.5-122B on Mac Studio 96GB: Fixed 3 bugs that made long-context inference usable https://www.reddit.com/r/LocalLLaMA/comments/1uuwrc0/running_qwen35122b_on_mac_studio_96gb_fixed_3/
144 2151 reddit INFRASTRUCTURE 0.050 Someone built an AI agent that hacks networks and holds data for ransom. It just worked. https://www.reddit.com/r/artificial/comments/1uuouu7/someone_built_an_ai_agent_that_hacks_networks_and/
145 2097 hackernews UNCATEGORIZED 0.050 The One-Step Trap (In AI Research) http://incompleteideas.net/IncIdeas/OneStepTrap.html
146 2086 hackernews BUSINESS 0.000 I love LLMs, I hate hype https://geohot.github.io//blog/jekyll/update/2026/07/12/i-love-llms.html
147 2225 hackernews INFRASTRUCTURE 0.570 Show HN: Juggler – an open-source GUI coding agent, by the creator of JUCE https://github.com/juggler-ai/juggler
148 2095 hackernews INFRASTRUCTURE 0.020 Mechanistic interpretability researchers applying causality theory to LLMs https://cacm.acm.org/news/can-we-understand-how-large-language-models-reason/
149 2445 reddit PROBLEM SOLVED 0.260 Ph.D. in Operations Research / Big Tech Eng: How to transition into intermediate/advanced ML for high-value industries (Robotics, Defense, Finance)? [D] https://www.reddit.com/r/MachineLearning/comments/1uumkkg/phd_in_operations_research_big_tech_eng_how_to/
150 2089 hackernews PROBLEM SOLVED 0.410 Migrating a production AI agent to GPT-5.6: 2.2x faster, 27% cheaper https://ploy.ai/blog/migrating-a-production-ai-agent-to-gpt-5-6
151 2320 reddit UNCATEGORIZED 0.010 this openai court story is starting to look ugly https://www.reddit.com/r/artificial/comments/1uul5ef/this_openai_court_story_is_starting_to_look_ugly/
152 2039 rss UNCATEGORIZED 0.000 LinkedIn is the undisputed king of long-form AI slop, according to a study spanning five platforms https://the-decoder.com/linkedin-is-the-undisputed-king-of-long-form-ai-slop-according-to-a-study-spanning-five-platforms/
153 2037 rss MODEL RELEASE 0.090 Claude Code now has a built-in browser that lets the AI read, click, and type on external websites https://the-decoder.com/claude-code-now-has-a-built-in-browser-that-lets-the-ai-read-click-and-type-on-external-websites/
154 1571 reddit SHIPPING 0.190 Kreuzberg (local document extraction) is being renamed to Xberg - current version on LTS https://www.reddit.com/r/LocalLLaMA/comments/1uuhqlz/kreuzberg_local_document_extraction_is_being/
155 1980 reddit UNCATEGORIZED 0.170 Local Image to 3D (<2gb RAM, <20s, Apple Silicon, iPhone) https://www.reddit.com/r/LocalLLaMA/comments/1uuga40/local_image_to_3d_2gb_ram_20s_apple_silicon_iphone/
156 2092 hackernews PROBLEM SOLVED 0.190 AI boosts research careers but narrow the span of ideas explored: study https://spectrum.ieee.org/ai-science-research-flattens-discovery
157 1976 reddit LOCAL AI 0.440 If you use Open Code or other agenting programs you are leaving a lot of t/s if you don't actually use agents in parallel. Benchmark : RTX5090, Qwen3.6 35B loaded via LM studio with parallel tasks set to 8 https://www.reddit.com/r/LocalLLaMA/comments/1uueuks/if_you_use_open_code_or_other_agenting_programs/
158 2427 reddit SHIPPING 0.300 Zer0Fit: I took Google's new TabFM & TimesFM ML foundation models and made them available as an MCP server for zero-shot ML tasks (forecasts / classifications / regressions). 100% local. [P] https://www.reddit.com/r/MachineLearning/comments/1uue8cc/zer0fit_i_took_googles_new_tabfm_timesfm_ml/
159 1573 reddit LOCAL AI 0.260 I got Nemotron Puzzle 75B running smoothly on a 64GB M2 Max https://www.reddit.com/r/LocalLLaMA/comments/1uue46z/i_got_nemotron_puzzle_75b_running_smoothly_on_a/
160 1984 reddit UNCATEGORIZED 0.050 Working around Qwen3.6-27B's tool-call failures and looping https://www.reddit.com/r/LocalLLaMA/comments/1uue278/working_around_qwen3627bs_toolcall_failures_and/
161 1965 reddit SHIPPING 0.260 Zer0Fit: I took Google's new TabFM & TimesFM ML foundation models and made them available as an MCP server for zero-shot ML tasks (forecasts / classifications / regressions). 100% local. https://www.reddit.com/r/LocalLLaMA/comments/1uudxi8/zer0fit_i_took_googles_new_tabfm_timesfm_ml/
162 2041 rss PROBLEM SOLVED 0.170 S&P Global sees OpenAI as a "key credit risk" for Oracle and cuts its credit rating https://the-decoder.com/sp-global-sees-openai-as-a-key-credit-risk-for-oracle-and-cuts-its-credit-rating/
163 2431 reddit UNCATEGORIZED 0.050 Obtaining Irregular Learning Curves with HyberBand Tuned ANN model for Price Prediction [P] https://www.reddit.com/r/MachineLearning/comments/1uud3qj/obtaining_irregular_learning_curves_with/
164 2042 rss CULTURE 0.030 Meta kills Muse Image feature that let anyone generate AI photos of Instagram users without consent https://the-decoder.com/meta-kills-muse-image-feature-that-let-anyone-generate-ai-photos-of-instagram-users-without-consent/
165 2087 hackernews INFRASTRUCTURE 0.130 Old and new apps, via modern coding agents https://terrytao.wordpress.com/2026/07/11/old-and-new-apps-via-modern-coding-agents/
166 1968 reddit PROBLEM SOLVED 0.260 Benchmark - 4x 5060 Ti (64GB VRAM) (P2P) - Qwen3.6 27B (INT8 /w bf16 kv cache) @ 8 concurrency with SGLang. SGLang seems to handle higher concurrency better with this setup https://www.reddit.com/r/LocalLLaMA/comments/1uuc3pi/benchmark_4x_5060_ti_64gb_vram_p2p_qwen36_27b/
167 1923 rss MODEL RELEASE 0.090 Claude Cowork's biggest use case is the mundane office work nobody wants to own, Anthropic says https://the-decoder.com/claude-coworks-biggest-use-case-is-the-mundane-office-work-nobody-wants-to-own-anthropic-says/
168 1921 rss CULTURE 0.000 OpenAI CEO Altman is now "pretty sure" AI is net job-creating, which is quite the pivot from predicting mass layoffs https://the-decoder.com/openai-ceo-altman-is-now-pretty-sure-ai-is-net-job-creating-which-is-quite-the-pivot-from-predicting-mass-layoffs/
169 1977 reddit LOCAL AI 0.450 Voodoo Quant beats Unsloth Dynamic 2.0 KLD by 95% in Qwen3.5 0.8B and 2B https://www.reddit.com/r/LocalLLaMA/comments/1uua3jd/voodoo_quant_beats_unsloth_dynamic_20_kld_by_95/
170 1924 rss UNCATEGORIZED 0.000 Grades dropped from 96 to 48 percent when a Brown professor made students take the exam without AI https://the-decoder.com/grades-dropped-from-96-to-48-percent-when-a-brown-professor-made-students-take-the-exam-without-ai/
171 1824 rss INFRASTRUCTURE 0.040 AI agents win at Slay the Spire 2 after researchers replace growing chat logs with structured memory https://the-decoder.com/ai-agents-win-at-slay-the-spire-2-after-researchers-replace-growing-chat-logs-with-structured-memory/
172 1564 reddit MODEL RELEASE 0.100 Need help tuning cache in llama-server https://www.reddit.com/r/LocalLLaMA/comments/1uu8g9f/need_help_tuning_cache_in_llamaserver/
173 2094 hackernews INFRASTRUCTURE 0.180 Show HN: Mindwalk – Replay coding-agent sessions on a 3D map of your codebase https://github.com/cosmtrek/mindwalk
174 1569 reddit PROBLEM SOLVED 0.180 i would like to share my experience. working with huge LLMs and poor Machine https://www.reddit.com/r/LocalLLaMA/comments/1uu6qvh/i_would_like_to_share_my_experience_working_with/
175 1975 reddit INFRASTRUCTURE 0.320 **Your $80 Tesla P100 has been doing silently noisy math in llama.cpp for years. Three lines fix it, for free.** https://www.reddit.com/r/LocalLLaMA/comments/1uu6p9o/your_80_tesla_p100_has_been_doing_silently_noisy/
176 1978 reddit UNCATEGORIZED 0.010 I mapped Anthropic’s J-Space Hallucination signal across 7 datasets on Qwen3-4B to find out where it works and where it breaks https://www.reddit.com/r/LocalLLaMA/comments/1uu61wb/i_mapped_anthropics_jspace_hallucination_signal/
177 1566 reddit INFRASTRUCTURE 0.120 First attempts at a CPU setup - MS-02 Intel 285hx, trying Qwen3, Qwen3.6 and Gemma4 https://www.reddit.com/r/LocalLLaMA/comments/1uu5ht0/first_attempts_at_a_cpu_setup_ms02_intel_285hx/
178 1966 reddit UNCATEGORIZED 0.010 I didn't give up - extGemma4-40_5B returned https://www.reddit.com/r/LocalLLaMA/comments/1uu4hxp/i_didnt_give_up_extgemma440_5b_returned/
179 1572 reddit LOCAL AI 0.300 Qwenthropic https://www.reddit.com/r/LocalLLaMA/comments/1uu3545/qwenthropic/
180 1983 reddit LOCAL AI 0.300 Interactive Jacobian-Lens visualizer and live steerer for GGUF models on llama.cpp https://www.reddit.com/r/LocalLLaMA/comments/1uu32z6/interactive_jacobianlens_visualizer_and_live/
181 2085 hackernews MODEL RELEASE 0.270 What xAI's Grok build CLI sends to xAI: A wire-level analysis https://gist.github.com/cereblab/dc9a40bc26120f4540e4e09b75ffb547
182 1565 reddit LOCAL AI 0.340 Measuring PCIe transfer under dual GPU with pipeline & tensor llama.cpp https://www.reddit.com/r/LocalLLaMA/comments/1utz50z/measuring_pcie_transfer_under_dual_gpu_with/
183 2088 hackernews UNCATEGORIZED 0.020 Mesh LLM: distributed AI computing on iroh https://www.iroh.computer/blog/mesh-llm
184 2090 hackernews UNCATEGORIZED 0.020 Stop Telling Me to Ask an LLM https://blog.yaelwrites.com/stop-telling-me-to-ask-an-llm/
185 1967 reddit PROBLEM SOLVED 0.180 Ultra budget 20GB vram with 448GB/s for $100 bucks. https://www.reddit.com/r/LocalLLaMA/comments/1utwqf8/ultra_budget_20gb_vram_with_448gbs_for_100_bucks/
186 1559 reddit INFRASTRUCTURE 0.340 Performance comparison on full compute performance (Anima) and LLM prompt processing of 5090 (600,475 and 400W) vs 6000 PRO MaxQ shunt modded and water cooled (at 300, 400, 475 and 600W), and 6000 PRO WS/SE (600W). https://www.reddit.com/r/LocalLLaMA/comments/1utvbey/performance_comparison_on_full_compute/
187 1558 reddit LOCAL AI 0.500 I benched quad 5060Tis for code generation with Qwen3.6-27B so you don't have to (it's really good) https://www.reddit.com/r/LocalLLaMA/comments/1uturng/i_benched_quad_5060tis_for_code_generation_with/
188 2096 hackernews BUSINESS 0.020 Wealthy AI workers send San Francisco house prices soaring https://www.bbc.com/news/articles/c9q29j47v9ro
189 1992 hackernews CULTURE 0.020 AI 2040 and the cult of intelligence https://geohot.github.io//blog/jekyll/update/2026/07/11/ai-2040.html
190 2000 hackernews INFRASTRUCTURE 0.100 Who manages the agents? https://www.off-policy.com/dont-go-quietly-into-the-ai-night/
191 1828 rss UNCATEGORIZED 0.000 OpenAI's GPT-5.6 Sol Ultra reportedly solves a 50-year-old math problem in under an hour https://the-decoder.com/openais-gpt-5-6-sol-ultra-reportedly-solves-a-50-year-old-math-problem-in-under-an-hour/
192 1999 hackernews CULTURE 0.020 Reverse centaurs are the answer to the AI paradox (2025) https://pluralistic.net/2025/09/11/vulgar-thatcherism/#there-is-an-alternative
193 1731 rss MODEL RELEASE 0.090 Terrorist groups are using every major AI chatbot for attack planning and weapons development https://the-decoder.com/terrorist-groups-are-using-every-major-ai-chatbot-for-attack-planning-and-weapons-development/
194 2001 hackernews INFRASTRUCTURE 0.180 Show HN: Reame – a CPU inference server that gets faster as it runs https://github.com/swellweb/reame
195 1730 rss BUSINESS 0.000 OpenAI bets on families as ChatGPT goes deeper into households https://techcrunch.com/2026/07/11/openai-bets-on-families-as-chatgpt-goes-deeper-into-households/
196 1776 hackernews UNCATEGORIZED 0.020 Ghost Font: A font that humans can read but AI cannot https://www.mixfont.com/ghost-font
197 1786 hackernews BUSINESS 0.020 Microsoft latest report shows 25% emissions raised due to AI data centers https://www.windowscentral.com/microsoft/dropping-greenwashing-credits-and-expanding-ai-datacenters-caused-microsofts-25-percent-emissions-jump
198 1592 hackernews PROBLEM SOLVED 0.190 Companies are scrambling to curtail soaring AI costs https://www.economist.com/business/2026/06/14/companies-are-scrambling-to-curtail-soaring-ai-costs
199 1591 hackernews SHIPPING 0.230 Meta pulls new AI image feature after days of backlash https://www.bbc.com/news/articles/c2dy6e8klw0o
200 1590 hackernews PROBLEM SOLVED 0.190 AI Can't Recreate the Thrust Game (But It Can Help You Understand It) https://www.jamesdrandall.com/posts/thrust_ai_powered_software_archaeology/
201 1588 hackernews CULTURE 0.020 Apple sues OpenAI, accusing it of stealing company secrets https://www.nytimes.com/2026/07/10/technology/apple-openai-lawsuit.html
render_site.py
+1 -1
View File
@@ -14,7 +14,7 @@ from collections import OrderedDict
import clickability as cb
HERE = os.path.dirname(os.path.abspath(__file__))
WEBROOT = "/var/www/preprod3"
WEBROOT = "/var/www/preprod2"
DB_PATH = os.path.join(HERE, "oracle.db")
TOP_N = 8
HALF_LIFE_H = 18.0
site/feed.json
+2044 -762
View File
File diff suppressed because it is too large Load Diff
site/index.html
+2261 -793
View File
File diff suppressed because it is too large Load Diff
today_yesterday_aind.tsv
+15
View File
@@ -0,0 +1,15 @@
rank id date bucket source title url aiND_tier lens_hit
1 2315 2026-07-14 SHIPPING reddit Open Source Local LLM Training Tool (for consumer hardware) https://www.reddit.com/r/artificial/comments/1uwcah2/open_source_local_llm_training_tool_for_consumer/ T1 real build
2 2461 2026-07-14 MODEL RELEASE hackernews Show HN: I RL-trained an agent that trains models with RL (for ~$1.3k) https://github.com/Danau5tin/ai-trains-ai T1 real build + cost
3 2648 2026-07-14 PROBLEM SOLVED reddit Developers Hate AI. I Used It To Sell 10 Websites This Week. https://www.reddit.com/r/artificial/comments/1uwj75g/developers_hate_ai_i_used_it_to_sell_10_websites/ T1 real outcome (paid)
4 2656 2026-07-14 SHIPPING reddit Structured output reliability with LLMs — 3-month production learnings https://www.reddit.com/r/artificial/comments/1uwe9qp/structured_output_reliability_with_llms_3month/ T5 real lesson
5 3166 2026-07-15 SHIPPING hackernews Open-source memory for coding agents, synced over SSH https://github.com/vshulcz/deja-vu/ T2 real OSS tool
6 3167 2026-07-15 UNCATEGORIZED hackernews Inkling – Open-Weights 975B Parameter LLM https://thinkingmachines.ai/inkling/ T2 real model release
7 2326 2026-07-14 INFRASTRUCTURE reddit I'm not a great artist — so I made an agent that turns my doodles into charcoal vectors https://www.reddit.com/r/artificial/comments/1uwbt7o/im_not_a_great_artist_so_i_made_an_agent_that/ T2 real build
8 2429 2026-07-14 MODEL RELEASE reddit [P] RL-training Qwen3.6 to RL-train tool using AI models [P] https://www.reddit.com/r/MachineLearning/comments/1uwfmfa/p_rltraining_qwen36_to_rltrain_tool_using_ai/ T1 real build
9 2438 2026-07-13 PROBLEM SOLVED reddit GPUHedge: Hedging serverless GPU providers improves cold start p95 117s->30s https://www.reddit.com/r/MachineLearning/comments/1uvlb6h/gpuhedge_hedging_serverless_gpu_providers/ T4 real measurement
10 2441 2026-07-14 PROBLEM SOLVED reddit New LLM Coordination Benchmark - Benchmarking Open-Ended Multi-Agent Coordination https://www.reddit.com/r/MachineLearning/comments/1uwc6ni/new_llm_coordination_benchmark_benchmarking/ T4 real benchmark
11 2640 2026-07-14 RESEARCH arxiv ViHoRec: A Quality-Controlled Vietnamese Hotel Recommendation Dataset + Cold-Start Benchmark https://arxiv.org/abs/2607.12946v1 T4 real dataset/benchmark
12 2638 2026-07-14 RESEARCH arxiv Form, Not Content? A Preregistered, Placebo-Controlled Evaluation of Self-Repair in Frozen Small Code Models https://arxiv.org/abs/2607.12962v1 T5 real lesson (eval)
13 2637 2026-07-14 RESEARCH arxiv The Illusion of Robustness: Aggregate Accuracy Hides Prediction Flips under Task-Irrelevant Context https://arxiv.org/abs/2607.12963v1 T5 real lesson
14 2735 2026-07-14 SHIPPING github quantumbyte: Open-source app builder engine — intent to workflow https://github.com/QuantumByteOSS/quantumbyte T2 real OSS build
1 rank id date bucket source title url aiND_tier lens_hit
2 1 2315 2026-07-14 SHIPPING reddit Open Source Local LLM Training Tool (for consumer hardware) https://www.reddit.com/r/artificial/comments/1uwcah2/open_source_local_llm_training_tool_for_consumer/ T1 real build
3 2 2461 2026-07-14 MODEL RELEASE hackernews Show HN: I RL-trained an agent that trains models with RL (for ~$1.3k) https://github.com/Danau5tin/ai-trains-ai T1 real build + cost
4 3 2648 2026-07-14 PROBLEM SOLVED reddit Developers Hate AI. I Used It To Sell 10 Websites This Week. https://www.reddit.com/r/artificial/comments/1uwj75g/developers_hate_ai_i_used_it_to_sell_10_websites/ T1 real outcome (paid)
5 4 2656 2026-07-14 SHIPPING reddit Structured output reliability with LLMs — 3-month production learnings https://www.reddit.com/r/artificial/comments/1uwe9qp/structured_output_reliability_with_llms_3month/ T5 real lesson
6 5 3166 2026-07-15 SHIPPING hackernews Open-source memory for coding agents, synced over SSH https://github.com/vshulcz/deja-vu/ T2 real OSS tool
7 6 3167 2026-07-15 UNCATEGORIZED hackernews Inkling – Open-Weights 975B Parameter LLM https://thinkingmachines.ai/inkling/ T2 real model release
8 7 2326 2026-07-14 INFRASTRUCTURE reddit I'm not a great artist — so I made an agent that turns my doodles into charcoal vectors https://www.reddit.com/r/artificial/comments/1uwbt7o/im_not_a_great_artist_so_i_made_an_agent_that/ T2 real build
9 8 2429 2026-07-14 MODEL RELEASE reddit [P] RL-training Qwen3.6 to RL-train tool using AI models [P] https://www.reddit.com/r/MachineLearning/comments/1uwfmfa/p_rltraining_qwen36_to_rltrain_tool_using_ai/ T1 real build
10 9 2438 2026-07-13 PROBLEM SOLVED reddit GPUHedge: Hedging serverless GPU providers improves cold start p95 117s->30s https://www.reddit.com/r/MachineLearning/comments/1uvlb6h/gpuhedge_hedging_serverless_gpu_providers/ T4 real measurement
11 10 2441 2026-07-14 PROBLEM SOLVED reddit New LLM Coordination Benchmark - Benchmarking Open-Ended Multi-Agent Coordination https://www.reddit.com/r/MachineLearning/comments/1uwc6ni/new_llm_coordination_benchmark_benchmarking/ T4 real benchmark
12 11 2640 2026-07-14 RESEARCH arxiv ViHoRec: A Quality-Controlled Vietnamese Hotel Recommendation Dataset + Cold-Start Benchmark https://arxiv.org/abs/2607.12946v1 T4 real dataset/benchmark
13 12 2638 2026-07-14 RESEARCH arxiv Form, Not Content? A Preregistered, Placebo-Controlled Evaluation of Self-Repair in Frozen Small Code Models https://arxiv.org/abs/2607.12962v1 T5 real lesson (eval)
14 13 2637 2026-07-14 RESEARCH arxiv The Illusion of Robustness: Aggregate Accuracy Hides Prediction Flips under Task-Irrelevant Context https://arxiv.org/abs/2607.12963v1 T5 real lesson
15 14 2735 2026-07-14 SHIPPING github quantumbyte: Open-source app builder engine — intent to workflow https://github.com/QuantumByteOSS/quantumbyte T2 real OSS build
transcripts/chunks/c00.mp3
BIN
View File
Binary file not shown.
transcripts/chunks/c00.txt
+106
View File
@@ -0,0 +1,106 @@
# oracle/transcripts/chunks/c00.mp3
# model: tiny | lang=en
[0.0s] A gented offering systems full one hour course,
[3.5s] how to build an automate anything.
[5.8s] So today I'm going to show you exactly how to build your own agent
[8.9s] offering a system based on what's working for me.
[11.0s] If you're wondering what an agent offering a system is,
[14.0s] basically it's a way to have all of your agents
[17.5s] inside one beautiful dashboard,
[20.2s] where you also build out custom workflows
[22.9s] and you have everything one click away.
[25.1s] And the amazing thing about this is
[27.4s] that you can basically have everything in one single place
[31.6s] that's ready to automate and offer it as you want.
[35.0s] So let me give an example.
[36.6s] For me, I suspend a lot of time create videos,
[39.3s] and so we create this video agent, the basically does it for us
[42.5s] in one single click.
[44.0s] We have, for example, an SEO content pipeline where we can research stuff,
[47.9s] we can find keywords who can generate content,
[50.2s] and then we just deploy to our website in one single click.
[53.3s] And basically everything that you want to automate,
[55.4s] you can automate using these systems.
[57.4s] So for example, Hermes agent, not very fine in the thermal.
[60.7s] If you're just using the thermal, not that great.
[62.7s] Even if you're using for example Hermes desktop,
[65.1s] you still can't get it exactly how you want.
[67.9s] With this system, for example, we have Hermes Jarvis, which is a voice-operated AI agent
[73.3s] and looks and feels exactly how I want it to look.
[75.7s] It's got all of my projects over here.
[78.1s] And what else I have, for example, at Hermes, I'll recall the pause
[80.5s] and let's trend it news and if it melts.
[82.1s] And we're not seeing that to both stores to show off my own own product.
[84.9s] I just want to show you what's possible.
[86.6s] With this sort of stuff, for example, this one,
[89.0s] this is Hermes Astros, which again, is a custom workflow.
[92.0s] And you can rename this, you can change it how you want,
[94.9s] you can build whatever you want using these systems.
[97.4s] And this, for example, actually takes the latest news from Twitter,
[101.2s] gives us the trending topic ideas.
[104.9s] Looks at different titles we could use for that content.
[108.3s] And then also we can just plug out to the video agent.
[110.8s] So it can automate the video.
[112.0s] We can plug it into notebook, lemon.
[113.5s] Create, for example, free infographics and all that sort of thing.
[116.1s] And then we could create a serial content for you.
[117.8s] And you can build all of this for free using your existing subscriptions as well,
[122.3s] which is very important to note.
[123.6s] So, for example, if you have chat cheapity already,
[126.4s] and you already subscribed to it,
[127.9s] well, you can actually use chat cheapity to build all this out.
[131.4s] I actually personally used Opus 418 inside,
[134.4s] call it an Avise Fable 5 to help me a lot along the way.
[137.0s] And for example, if you have something that's a problem you want to solve,
[141.4s] let me give an example.
[142.6s] So a lot of you ask me like, how do you automate leads and email outreach using something like
[147.1s] Hermes?
[147.8s] So with Hermes, we have a tool here.
[149.6s] The congenerate leads, we just type in exactly what type of leads you want to get here.
[154.2s] It generates a list leads.
[155.8s] Then it has all the campaigns.
[157.2s] It can write the email campaign itself.
[159.3s] Then it can actually send the emails with these campaigns,
[161.9s] it pulls in all the lead details.
[163.7s] And then from there, we can manage the metrics.
[165.7s] We can see how many leads we've got, how many we've validated,
[168.2s] how many ascendable, how many we actually send, how many replied,
[171.1s] and we've got our inbox and sent inbox fully automated with Hermes agent,
[175.8s] it's like one beautiful dashboard.
[177.4s] And that just saves so much time.
[179.7s] So this is the power of this stuff.
[181.4s] This is how amazing it is.
[184.0s] And you will learn how to build all this today.
[187.0s] And I'm going to show you exactly how we can get started.
[190.4s] So if you're learning why I'm the right person to show you this,
[195.3s] I literally spent about three to four hours a day
[198.7s] coding this out and making it better and better.
[200.5s] And there's so many mistakes that I learned along the way,
[202.4s] so many things that I can prove.
[203.8s] So with that, let's get straight into this.
[206.8s] So I'm just going to be using G55.6 as an example.
[209.6s] I quite like soul.
[210.6s] I think it's great to build with.
[212.2s] And you can see that we've got GVT and chatGVT work over here.
[216.5s] It's sort of snap.
[217.7s] Now for example, we set up Hermes agent.
[220.6s] That's fully ready to go with use GVT.5.6 as a brain.
[225.3s] Inside Hermes agent and now we want to say create a dashboard with it.
[229.8s] So what we can do here is we can actually go inside the chat here and we can say, okay,
[235.6s] now create a mission control dashboard where I can operate chatGVT,
[244.6s] codex plus Hermes agent.
[249.5s] And we can say make it clean, make it beautiful, make it interested to use
[255.4s] make sure it actually works.
[257.0s] And then if you have an idea of what you want to create,
[259.8s] so let's say for example, we have a certain design style with that sort of thing,
[262.7s] we can actually use that for inspiration when you're building this out.
[266.6s] So if we wanted to build this setup inside chatGVT,
[270.6s] we can just take screenshot, plug in and just say,
[274.1s] I've attached an example of the style I wanted to create.
[278.5s] And so we can hit enter now and that'll begin building this out.
[283.4s] So if you want to start from zero and you want to build it exactly how you would like it,
[289.5s] in a way that's easy to use, you can go from there.
[292.4s] That'll also show you another way that you can do this.
[294.6s] So if you want to use someone else's template,
[297.8s] so for example, we have a
transcripts/chunks/c01.mp3
BIN
View File
Binary file not shown.
transcripts/chunks/c01.txt
+53
View File
@@ -0,0 +1,53 @@
# oracle/transcripts/chunks/c01.mp3
# model: tiny | lang=en
[0.0s] version of mine inside there, but bottom that we actually update every time I code out for
[4.8s] three or four hours, I will actually save that. So you can see the sit file here. If you
[9.8s] want to do it, how do you install this? See we do the same thing. You can use whatever agent
[15.3s] you prefer so that could be checked with the it could be for example called whatever you prefer.
[20.3s] And we can start a new project. Start from scratch, save that new project. And then we can
[26.7s] add this so we can add the sit file. So we can grab the sit file over here, which is
[34.7s] there. And we can say install and configure this agent or S on a local port we don't already
[43.4s] have running. So I like to run these things locally as well. And we'll come onto that and
[48.2s] why in a second. But now we have two options. So we can use the setup from the Air Profit
[53.7s] for the bottom and that's just kind of like ready to go and all the other option that
[57.6s] we have is if you just want to build this out free from scratch you can go and build it inside
[62.9s] chat to you with you as you can see. So now it's going to say okay what should the dashboard be
[68.1s] able to do. So says before implementation, I need three decisions that material effect the
[74.8s] architecture. So we have Asian control. We have workspace. We have application format. Now for
[81.1s] me, I'm going to say just allow it to run locally for me in Chrome. And have one fix to
[90.4s] folder and allow it to be fully operational. So this is something important to note. CC when it's
[96.2s] asking you it's asking you like do you want it for operations so that it can actually like edit
[99.6s] files. It can do things autonomously for you or do you want it kind of like a chat. For me, I
[104.8s] don't want a chat because if I just want a chat I'll just go to chat to chat to be here.
[109.8s] And that is exactly why I actually created the agent OS because when you're just going back and
[116.1s] forth inside the chat here, it's super limited. Like it doesn't feel authentic, it doesn't feel
[120.1s] powerful. It doesn't really have everything you want and it's very, very limited in terms of what
[124.8s] it could do. So that's why I prefer to be like way more powerful. The whole point of this is to build
[128.9s] something absolutely amazing that can be a little bit of an elephant and just runs like magic. So that's
[134.2s] why we are setting up like this system. So you can see over here this is beginning to install now as well.
[142.9s] Alright so you can see that we've got two options here. We have the agent OS been installed locally from
[147.6s] template and we have the Hermes agent OS over here too. See two different options. Now both of these
[152.8s] will run like you can see I actually want to switch this to fast mode. So I'm going to say switch to fast mode now.
[159.1s] Just so that we can get faster outputs from both of these. That's already enabled which is great.
[164.4s] And now when we're using here you can see that it's beginning to use this local setup. So you'll see
[172.0s] a number like this. Now you might be wondering why you're running it locally, why not just run it on VPS.
[177.5s] Here's something to consider. Okay and this is something that I saw when I was looking at all in the
[182.5s] stakes from open core and a lot of the issues that people face is that there were two things with agent. So you
[188.2s] have to be very careful of. Number one is that you have to be careful because when you're building
[194.3s] with this sort of stuff, if you give access to multiple different devices then someone accidentally could
[202.2s] access your agent OS and if it can edit files if it can control your computer if you can do everything like that
[207.9s] it's not that secure. If you if you've given the world access to it. So that's why I don't recommend that. Instead
[213.9s] what I recommend is that you use the system like this and you run it locally. Now you can run it on
[219.4s] VPS. We have plenty of AR profitable and members who do that and then they just limit the tools that
[224.2s] they give access to but it's up to you how you want to do it. Now you can see that we already have the agent
[228.9s] OS installed over here. So this is fully installed as you can see here. We've got a friend setup. It's set up on a
[236.0s] separate agent OS local port. So for example if we go over here this is my actual agent OS and then we have
[243.6s] the template over here. So it's pretty easy and the thing to note here is when you're using this you just need to
[248.8s] configure everything. So it's now running using the template that gave it but I'm going to say configure it
[255.7s] here. I'm just going to make this simple if you're doing the sound. We're going to rename this project and we're
[260.8s] going to call that template setup. And then we're going to use this project here and we're going to call
[268.8s] starting from scratch. We'll just call this new agent OS example to a two different options right there.
[279.6s] So now it is said okay before I configure it just we're just going to go through step by step.
[288.9s] Set it up at each part that doesn't require an API and we're going to say yes and figure it up.
[294.7s] Right just so that it can start working directly now we have
transcripts/chunks/c02.mp3
BIN
View File
Binary file not shown.
transcripts/chunks/c02.txt
+35
View File
@@ -0,0 +1,35 @@
# oracle/transcripts/chunks/c02.mp3
# model: tiny | lang=en
[0.5s] the other setup here. So this is the upper version working locally. As you can see and then I just want to make sure one thing as well
[8.7s] that we don't have overlap in ports. Yeah, so this is
[12.1s] starting on this port which is 3,200 and this one is running on that port which is 1,200. Now you actually see when you're doing this like it does a very good job of
[24.4s] building
[25.4s] how locally and also emulating the style that we want. So we'll see how this goes.
[34.0s] So it says course, scaffolding is in. I'm wearing this out. Now while we went for those two tasks both the new one and the template version to be done,
[44.5s] something else to consider is the memory setup. Now there's so many different memory options.
[50.2s] For example, if you've never come across it, Hermes agent memory, so many different options for that. You can use loads of different assistant memory. There's lots of different options for this.
[61.2s] The reason that you would have a memory is so that you're not starting from scratch and also that you'd link everything together.
[67.2s] So if we have a look for example at this setup over here. When you're building this out, all of these agents can use the context from our memory system. Why would we want that?
[80.2s] We would want that because for example, one of the reasons you want to have an agent or S is that you don't go in between different apps and have to configure everything separately.
[88.2s] So normally what happens is this is a big problem for most people is like they will go over to chat,
[95.2s] to chat, go over to chat, and then go over to the cloud and these two things they don't think together. They don't work together. So what you want it said is a memory system. How do you build a memory system? You can go into Obsidium Obsidium is my preferred memory system.
[110.2s] I'm sitting in basically stores everything in my town files in a beautifully organized system like this.
[116.2s] Now if you're wondering how to get Obsidium, you can just go to Obsidium and it's a free app that you can download. It's actually open source.
[124.2s] So you can just get Obsidium and then you configure it like that. It won't look as beautiful like this, but you can ask the agents to organize and smart and help you and organize it properly so it looks more efficient.
[138.2s] So step one is building out the dashboard which we've got chatchipity working on over here. Then step number two is that we are building the memory system inside all of this.
[150.2s] So that when we're using cloud and when we're using chatchipity and if we're using hermys they all sink together. They all work together.
[159.2s] So the reason for that is that then the agent is no exactly what you've worked on which means you'll get better responses. You'll get more personal responses and it'll be easier to use each one.
[170.2s] So if we have a look at our memory system over here and we scroll in this is our memory galaxy. So every single dot that you can see is a individual memory and this is all linked together inside one beautiful galaxy.
[184.2s] Where we've got all of our files over here that we can preview and we can have a look at really easy to use.
[191.2s] So when we're doing this it means that we can sink our memory together and we can have every agent basically understanding what the operations have worked on.
[202.2s] And the other thing to note about all of this system is like everything all dot inside here is for example like a person or a project or a business or a tool that I've used and it all gets sinked inside one place.
[215.2s] Now when you're doing this swap you don't have to create all of this yourself you should get your agents to do it. So you can say to for example, Claude, hey Claude, based on everything I've worked on recently just create an obsidian file and organize that and then you can say to chatchipity.
[231.2s] Based on my local obsidian file can you just improve this but that's how it works.
[236.2s] And one thing I'm just going to do here is I'm going to say don't remove the previous system that was set up on the local port just create a new one.
[247.2s] So we still need the agent or else on the old port because what it's done is basically take a mile port out and then replace it.
[254.2s] So make sure the old one on that's famous this local port is back, it's back up ASAP and then you can use it totally different local host.
[266.2s] So that's something to note as well as like when you're running this locally if you're running in other projects like other get her projects in the background.
[273.2s] You don't want to have them conflicting because you can only run one local project on one local year.
[282.2s] So we've talked about memory and we've talked about how to do dashboard now also it's important to note with the memory as well something that'll make it a lot easier.
[291.2s] You can use something like omie and omie can actually record your screen take notes can listen to your microphone.
transcripts/chunks/c03.mp3
BIN
View File
Binary file not shown.
transcripts/chunks/c03.txt
+76
View File
@@ -0,0 +1,76 @@
# oracle/transcripts/chunks/c03.mp3
# model: tiny | lang=en
[0.0s] phone, you can understand what you're working on and then it can create a list of memories
[3.7s] here.
[4.7s] So you can see for example, 21 minutes ago, one hour ago, one hour ago, it understands
[8.8s] everything that will work on recently.
[11.4s] That's really useful because then I don't have to stay at the nose myself and actually
[14.6s] get more detail than any of the other things that I'm working with.
[18.1s] So when we're using OME, which again is another free open source project, if you think
[23.6s] they do have a subscription if you're using it a lot, but if you're just using a basic version
[28.8s] of it, you can get access right here.
[30.8s] It's available on OME, it's MIT license, it blisters for you, understands what you're
[36.9s] working on, takes notes, and then you can actually export that to Obsidian.
[41.7s] So what OME, that goes to Obsidian and we go from there.
[46.0s] So if we go to Home inside OME and then we go to the settings and we check this out, we can then export
[57.8s] this directly to Obsidian as well, which is really, really useful.
[63.1s] So this is being an out now.
[65.3s] You'll see here, this is quite a big project for chat activities to work on.
[72.4s] So when it's being an out from scratch, you can see here that's beginning to do it out, said,
[76.2s] I said, I'm working on FIM, prioritizing the UI and the integrations here.
[85.2s] But it might take a few minutes, it's been coding out for like 10 minutes already, just working
[89.2s] on this brand new H&OS.
[91.0s] Whereas for example, as you saw before, we've already got the H&OS live with the templates, so it's
[97.4s] way faster to use the template than it is to use a brand new set of from scratch.
[103.8s] So if you want to save time, I would just go with the template inside the app, I'll put it in,
[107.3s] but if you want to build your own from scratch, this is how you can do it over here.
[111.0s] Now when we're waiting for those to run, one other thing to note here is that once we're
[116.8s] building this and we're creating stuff out, you only need to add what's relevant.
[122.8s] One feature that I built in and I recommend this for you too, is that you have an option to hide
[127.9s] things, because what we're trying to build here is something where we focus on the system,
[133.1s] not the model.
[134.1s] So for example, OpenCore, go back like two months ago, this was probably the biggest A-A-A-Gene
[140.5s] in the world.
[141.5s] It's only recently that people stop using it, and that includes me because it's Hermes to
[144.7s] go over.
[145.7s] So my point here is that you want to focus on the model, on the system, not the model.
[153.3s] So you can hide OpenCore and you can add it in.
[157.2s] And I would allow that for yourself as well, so that you can move things around, you can change
[161.2s] it, because things in AI are crazy.
[164.0s] Like G556, as soon as it came out, absolutely mind blowing.
[168.2s] But G555, I would never build anything like this with it.
[172.0s] So as you go along here, there's going to be things that you swap out that you don't need.
[177.0s] I mean, if you look at my setup as well, because I'm testing some of the things all the time,
[181.7s] I have loads of tools here, but for you personally, you might want to use Hermes or you might
[186.1s] only use Clawd and you just need to build out several workflows of each one.
[189.8s] So that's something important to note, it's like you don't have to use everything that you
[193.9s] see here.
[194.9s] It doesn't have to be as complex as this.
[196.8s] One thing that I learned along the way was the simpler you can make this the better, because
[203.2s] you won't use every feature.
[204.2s] For example, Kimmy code, I personally haven't used for weeks, so I should probably hide them.
[210.3s] I can always reuse it if Kimmy K-Table and A comes out and then it's absolutely mind blowing.
[216.2s] So you see how you can swap things in, you can swap things out, etc.
[220.1s] You've got the mission control here.
[223.1s] The daily goals list as well.
[225.2s] This is really useful.
[226.2s] I'd recommend building this into.
[227.6s] In any feature that you see here, that you like right, I want to add it in, or for example, let's
[232.0s] say you have an idea and you like I need to add that in, I need to add this in, etc.
[235.8s] Well, when you're building any of this out, you can go back into chatchipathy and you can say,
[240.6s] okay, great.
[241.9s] Now build out a daily journal section or now build out a daily habit tracker.
[248.5s] What are the features you have in your mind?
[251.2s] You can build in.
[253.2s] And you can ask chatchipathy to build in.
[255.4s] And if it's taken ages to reply, you can just say, okay, add a journal feature.
[262.2s] And if you type forward slash steer inside or if you add the message here, it will actually give
[269.1s] you the option to steer it in the right direction if it's in the middle of coding.
[273.7s] So that's how it works.
[274.7s] I said by said, now if you have a look, we've got the mission control template installed.
[281.2s] So this is actually the template from the AAPRF boarder.
[285.4s] If we have a look at the setup from the agenda, go S, the chatchipathy is built.
[290.8s] It looks similar, but this is the version where you're signed from scratch.
[295.6s] Now when you build an inside set by set, actually looks really nice.
[299.6s] When you build an inside set by set, actually looks really nice.
transcripts/chunks/c04.mp3
BIN
View File
Binary file not shown.
transcripts/chunks/c04.txt
+104
View File
@@ -0,0 +1,104 @@
# oracle/transcripts/chunks/c04.mp3
# model: tiny | lang=en
[0.0s] I said by said, you want to test it out.
[2.6s] So let's just test, are you working?
[5.2s] And you'll see things that you want to improve.
[6.6s] You'll see things that you want to make better.
[8.2s] And actually, if you have a look at this, it totally blanks.
[11.0s] So we said, are you working?
[12.9s] And then it just didn't reply.
[14.6s] So when we fix things to the flight, this and this way,
[17.5s] it can take quite a few hours per day
[19.6s] for me to test out and improve this.
[21.7s] We need to drag that screenshot in and say, hey,
[23.9s] this isn't working.
[25.3s] Can you fix it?
[27.4s] And we'll delete that previous response of the gig.
[30.0s] For example, is not replying and also when I click enter,
[34.4s] actually just had a blank screen.
[36.1s] So just make sure you test and actually make sure it works.
[38.6s] So it's much easier to start from a template,
[41.5s] a template, why actually work, whereas, for example, this set up one.
[44.6s] And you see how this is running through each stage of the process,
[48.8s] whereas, for example, with the new version from scratch,
[52.1s] it created a nice UI, but it didn't even test it,
[55.0s] even though we previously asked it to.
[56.4s] And that's on 2.5.6, so extra high.
[59.1s] So because it's such a big project and because it was coding for so long,
[62.2s] you can't have forgotten to do what we asked it to do.
[64.5s] Because we said, make sure it works.
[67.1s] Make sure it actually, you know, actually works,
[70.6s] set by set, but it doesn't, it just created the UI.
[73.0s] So with what the UI, the user interface that's looking good,
[77.0s] we just need to make sure that the rest of it is working to.
[81.0s] So it's going to start improving that as well.
[83.4s] So if we come back to scratch, you'd be fine.
[85.5s] That's beginning to analyze it as well.
[87.8s] Now also the great thing about chatchipathy,
[89.6s] and you can use code for this, you could even use her.
[91.7s] It's for it, although I think if I had a choice out of all these,
[94.3s] I'd use chatchipathy or code at this point,
[97.8s] because it's just much more visual when you build in stuff like this,
[100.6s] you can actually fork the conversation.
[103.4s] So you can click on new over here, and inside the same folder,
[108.4s] you can have agents working in parallel on building out the H&WS.
[112.6s] Now, this is really, really useful.
[114.6s] So for example, when I'm coding out the H&WS,
[117.8s] inside, called what you see on the left hand side
[121.3s] is that I will test and run different implementations side by side at the same time.
[126.4s] What I mean, for example, is that when I was building out the video agent,
[130.4s] I was also working on the H&WS.
[132.2s] And I was having them both being built side by side
[135.0s] inside the H&WS system.
[137.5s] And so the great thing about chatchipathy as well,
[141.3s] is that you can just start a new chat here,
[143.3s] and you can say add inside the H&WS,
[149.0s] an image generator.
[152.6s] And then we'll say use G556 for generating the images
[158.3s] with the OAuth slash CLI setup,
[162.8s] and make sure it actually works, but it's allow me to save anything that I create.
[167.7s] And so what you can see here is that we now have
[171.8s] two conversations at the same time
[174.2s] for our new H&WS that were built in.
[176.6s] So every idea that you have that you think,
[179.6s] wow, this would be valuable, you can build it at the same time.
[182.6s] And this means for example, you can have like six different conversations
[186.4s] inside the same project that would be encoded outside by side.
[189.8s] So we can have, for example, a video agent,
[192.8s] an SEO agent being coded out, the lead generation being coded out,
[196.1s] and we just add in new workflows inside the H&WS system,
[199.6s] as we go along.
[200.9s] So you can see here that it's asking for your questions,
[203.8s] I'm just going to say, yeah, just figure it out and build out, right?
[207.6s] I don't need to give instructions like that to G556,
[213.7s] but this is smart enough to figure out for me,
[215.5s] and if it really is bad then, okay, no problem.
[217.8s] I can go back and fix it later.
[220.2s] The other cool thing about using this is like,
[222.1s] whether you're using, for example, called or using chatGpc,
[225.0s] you can plug in your existing subscriptions into this system.
[229.3s] So if you use Claude Woll, ask it to build,
[232.2s] Claude CLI into the H&WS, and that way you don't need to use API,
[236.4s] which means that you don't blast through loads of tokens,
[238.7s] which means that it doesn't cost you anything extra,
[241.2s] to number one build the H&WS and number two use the H&WS.
[245.7s] Because for example, we have a look inside the system,
[248.9s] we have an agent who goes with Claude with Hermits,
[252.6s] with Vantigravity with Codex.
[254.2s] Now Codex already subscribed to chatGpc before I build this,
[257.2s] and so it didn't cost me anything extra.
[259.0s] Claude already subscribed to it before I build it,
[261.5s] didn't cost me anything extra.
[262.7s] Hermits free open source projects, we plug on the Rue sub-brain into it,
[266.6s] doesn't cost me anything.
[267.8s] So you see how you can build these agente cooperatives systems for free,
[271.1s] that's very important to note as well.
[273.7s] And now this is beginning to code step by step.
[278.4s] It's adding the image generator inside the existing storming home,
[281.4s] this is one.
[282.6s] And by the way, for the actual agent or system that uses the template,
[287.4s] that was all set up and configured here.
[289.3s] So it says the new system has separate configuration
[292.3s] and what I would do if you are installing this template from scratch on the upper floor,
[296.2s] it's just tell it, listen, test everything out, anything that doesn't.
transcripts/chunks/c05.mp3
BIN
View File
Binary file not shown.
transcripts/chunks/c05.txt
+78
View File
@@ -0,0 +1,78 @@
# oracle/transcripts/chunks/c05.mp3
# model: tiny | lang=en
[0.0s] work, tell me and make sure that you authenticate it or if you can't authenticate it,
[5.4s] that you ask me so that I can set out.
[7.8s] So the template set up is pretty much finished at that point.
[10.4s] That's way, way easier.
[11.7s] Whereas for example, be on an agent or a search, you know, you could be here for a few hours
[16.0s] or you know, you might see something that breaks tomorrow and then you have to go and fix it
[19.4s] as well.
[21.6s] Now, when we were talking about the memory system as well, that obsidian system, you can
[26.3s] go back into the agent or else here.
[29.7s] And you could say, also add the memory system from obsidian, so I have obsidian installed
[35.1s] locally and install that into the agent or else as well.
[38.5s] So it's automatically synced and also make sure that the chat activity and hermys are synced
[45.0s] to my memory system.
[46.5s] So when I use them, you update the obsidian as an immatarver.
[51.7s] But also when I ask you questions, you learn from the obsidian vault that we have locally
[57.4s] and you use that as context.
[60.0s] That way everything is personalized inside the H&O S.
[64.0s] So this is the memory system and how you can install that.
[67.6s] And this is pretty cool because now we have the front end plus functionality that's being tested
[75.2s] of it.
[76.2s] We have the custom workflow that we're building out here and we have the memory system that's
[82.8s] being built over here.
[84.4s] So we're adding three different layers as we go along.
[88.2s] We've done that in a space of 27 minutes.
[92.9s] There's going to be some workflows as well over the other ideas for and you're like, there's
[96.2s] no way it could do that.
[97.6s] I would genuinely just ask because most of the time, G556 can handle anything that you've
[103.1s] got out of here.
[104.1s] I've not seen G556 be unable to build this stuff out.
[108.5s] It's pretty unbelievable, everything that you can use here.
[113.1s] So I'm going to say figure this out.
[115.8s] If I had more time if I was here for a few hours or whatever, then what I would actually do is give
[121.4s] it more instructions and be more in-depth and just take a little bit longer to answer these
[126.5s] questions, but I'm trying to save you time and I'm trying to show you in real time how it works
[130.5s] about it being boring, honestly.
[132.6s] So if we have a look over here, we've now set up G556 so we've in-miss generation, we've set up
[138.9s] the front and functionality and we have the dashboard here.
[142.0s] We can also improve it and test out side by side inside chatchubity.
[146.8s] So this is the offering system of built and this is the dashboard that we're built.
[151.8s] So if we go inside the Hermiss agent here, we're just going to say working and make sure Hermiss
[157.5s] agent actually replies to us, boom, and it says yes, I'm working what do you need?
[161.9s] Now this is amazing because now we have Hermiss agent, we have G556 plugged in which is like
[168.0s] one of the best frontier models in the world.
[169.9s] We have chatchubity codex and that's creating some Missist stuff as well.
[175.6s] So if we go to Hermiss agent here, we can use it identically and we can use it inside our H&WS.
[180.8s] Now we've got the inside chatchubity codex and now we have the mission control as well.
[184.7s] If we click on any of these, it takes a straight to the commissation.
[188.0s] So UI looks nice and looks clean and looks simple, which is great as well.
[191.8s] And then over here we have the live activity, so we can see what's being done.
[197.0s] We have the work that moves with recent operations, we have both of the agents that we've built,
[201.4s] and we have everything else ready to go.
[204.6s] Unbelievable how easy it was to set this up.
[207.3s] Now of course we do have the emiss generator that needs to be tested and we'll see how that forms
[211.9s] as well.
[212.9s] And then we've also got the memory system that's being built out as well.
[217.8s] There's a lot of layers do you need to add to this as well.
[220.7s] That might be another question they're asking.
[222.8s] So something that's really, really useful is the conversation history.
[228.4s] Now if I open this up and it just starts from blank every time, not very useful.
[233.0s] But if I can start a new chat, like you can see here, and we can switch between them, that is great.
[238.6s] Because then you're kind of using Hermiss like chatchubity.
[240.8s] So you can go back and forth inside the chat of it, but also you can use it identically because
[246.8s] you've got Hermiss agent, basically have superpowers, you can do computers, you can edit files locally.
[253.0s] And so it's kind of like you've created your own better version of chatchubity, this is chatchubity,
[257.9s] but it can't edit my local files.
[259.5s] Whereas for example if we go over to Hermiss agent, this can work directly here too.
[263.2s] So that's great.
[263.9s] Now one thing that I also add here as well is a system where you can essentially save everything
[272.8s] that you've created.
[273.8s] Now come on to that in a second, but let's have a look at this image generator.
[277.8s] And you see how it would basically build an outer and soft tool.
[281.5s] Think about how amazing that is.
[283.5s] We're at the stage now where you can build your own tools, you don't need any technical knowledge,
[288.8s] you don't need to be really technical or understand HTML or be a coder to build this stuff out.
[295.1s] You can just build your own custom workflows using our automation in a way that is
transcripts/chunks/c06.mp3
BIN
View File
Binary file not shown.
transcripts/chunks/c06.txt
+54
View File
@@ -0,0 +1,54 @@
# oracle/transcripts/chunks/c06.mp3
# model: tiny | lang=en
[0.0s] custom-wise and personalized here. That is unbelievable. So if we go inside here and we're like,
[4.3s] okay, create a image of a cat eating biscuits and we'll delete this bit here, we'll change that.
[11.8s] We'll get rid of that. You see how it's got this aspect ratio as well? So I didn't even ask you to do that,
[17.0s] but code X4 ahead because that asks it to figure out. That's why I quite often prefer just
[21.4s] giving the AI to figure out because it has better ideas and it has more time to give attention to this
[26.7s] than I do. Like if I've been coding this out for hours, I'm going to get tired, I'm going to get
[32.2s] distracted, is going to be difficult to focus at times. You might get frustrated sometimes.
[37.1s] Well, where's if you give the decisions to a frontier model like GPT-5 on 6, it can do it all for
[43.2s] you and handle it as you go along. So now we can say, okay, I want to just go with a cinematic image of a
[49.6s] cat eating biscuits. We've got the prompt, we've got the aspect ratio with cinematic, we'll hit generate
[54.7s] image and that may not work, but so far it's actually looking really good, really impressive actually.
[61.5s] And so one thing that might happen here is if we click from image to video to homeization, I wonder if
[68.0s] it's going to stop? Yeah, see how we click on that and then it's stopped generating. So we need it to be
[73.0s] able to run in the background. That's another great feature. And you'll see the more you test it, the more you'll
[77.4s] see ideas from improve it and more things that will come up that are kind of like bugs by this. And so
[83.1s] what we want to do here is just say, okay, inside the custom workflow image generator and this is exactly
[89.2s] why you want to rename the conversation. So if you right click and rename the conversation, you can see exactly
[93.9s] what you're working on. It keeps makes it easy to keep track. We can go inside here and say, based on the image
[100.2s] you, if I generate an image, can you just make sure it still runs in the background whilst I'm moving around?
[107.1s] And also what we've seen here is that it's actually tested out and it's found some bugs that we didn't
[112.6s] even know about. So it's turned a 500 error and it's uncovered a real bug there which is fantastic
[119.4s] as well because it's still in the bug checking for me. But overall it's looking very, very promising
[125.0s] that's beginning to work. I mean, a bit of we go back to the app here and we're like, okay,
[131.8s] create an image of a cat and then we just hit generate image. That should actually work without us
[137.8s] switching between them. But let's see. So the mission control, we have the different agents built in.
[143.0s] We have the studio of custom workflows as well. But this is beginning to look and form something
[149.8s] that is actually amazing. And you can see here that we've got the agentic memory of all as well.
[157.0s] So this is the system that we've used and as you want to add that as a tab. So I'm going to say,
[163.0s] add the memory system as a tab. So the I can easily see the memory has been added and see what's
[170.0s] been done and also visualized the inside like a beautiful knowledge graph and we can hit into here.
[174.6s] Now it's beginning to run out of work usage as you can see because I've been running on fast mode and we've
[179.8s] created so much stuff started by side. I'm going to say figure this out. But the 8020 of building the
[185.5s] agentic operating system is being done as you can see. And so all of this is working in the background.
[190.5s] You do need to be careful with the token usage as well because you can easily run out of tokens.
[194.6s] If you're wondering how to check your token usage, you can go inside the advanced section here and you can
[200.6s] switch the effort levels and speed and everything else. As you can see how many tokens you've used side by side.
[207.4s] And you can also click on reset usage as well as an option. Now this is beginning to code out each part as well. Nice.
[217.0s] Memory system should be added in and we hit the limit. So there we go. We hit the usage limit already.
[226.2s] We can actually click on reset usage. I'll be quite interested. I know if we can and I'll. Oh yeah we can.
[231.2s] Of course you can do a weekly reset usage limit as well. I didn't know that.
[237.0s] That's pretty good. And then we'll go out weekly usage limit that resets on the 18th which is quite a few days ago.
[242.6s] But yeah and also when you're being out with this. If you want to reduce the amount of tokens
[248.6s] because this can be come up problem especially if you're creating some of the things in parallel using the
[254.4s] GenteColoring system, then you've got a few options. So you can switch your fast mode. You can switch off the
[259.9s] intelligence so you can actually dial down the intelligence level. You can also change and click on advanced
[266.8s] and then you can switch between the models. So if it's just something really basic then you could switch
[272.0s] the model to like 25.5 and that would use less tokens. You can change the effort level to light or medium
[279.0s] and you can change the speed standard. Now when you're building all this out,
[283.4s] the more you reduce the amount of tokens you use by dialing down the effort dialing down the speed,
[288.7s] dialing down the model, the worse the outputs will be. As always that balance with token optimization
[294.6s] creating something that you really like also using the plan.
transcripts/chunks/c07.mp3
BIN
View File
Binary file not shown.
transcripts/chunks/c07.txt
+115
View File
@@ -0,0 +1,115 @@
# oracle/transcripts/chunks/c07.mp3
# model: tiny | lang=en
[0.0s] about running out of credits.
[1.7s] Also, something else that's taken to account here
[4.2s] is that there are loads of different ways
[5.9s] you can reduce the amount of tokens and minimize them
[8.5s] even more.
[9.4s] We've actually got 10 different strategies
[11.1s] inside our token minimization playbook
[13.2s] in the app off of volume.
[14.3s] They can help you save like Nibes,
[15.7s] 95% of your tokens.
[16.9s] So if you want to get almost two to three X,
[19.6s] out of your system and be able to code for more
[23.7s] and use more tokens, then you can use this system here
[26.4s] and it will actually reduce the amount of tokens you use.
[29.2s] And that's also one of the reasons
[30.6s] that we recommend a template to set up
[32.6s] because for example, with the template to set up
[34.5s] as you can see right here, that just installed it,
[36.6s] configure it in the background,
[38.0s] and that was already done in like 10 to 20 minutes.
[40.4s] That is way better than having to build this from scratch yourself.
[44.4s] So it's your choice, but for me personally,
[47.8s] I would prefer to use like a template and set up
[49.9s] just so I can save time because, you know, times is really valuable.
[54.0s] So we have talked about how to build your own agent operating system,
[57.6s] how to add custom workflows,
[59.6s] how to set up the memory inside there,
[62.4s] how to test different works,
[63.9s] how to add new features, and you can see, for example,
[65.9s] we have the image of the cat E in this kit
[68.0s] now as well, so that custom workflow works really well.
[71.3s] One thing that I'd recommend to you all at your time,
[73.7s] so you look at where you spend your time
[75.7s] and then how can the agent OS help you?
[78.4s] So anything that you have an idea for,
[80.5s] you can build it as you see.
[81.9s] So for example, the image studio, we have that idea,
[84.2s] and we build it out.
[85.8s] Now at this point, you might be wondering, okay,
[87.9s] what should you be folks, you know,
[89.4s] what custom workflow should you add to your agent OS?
[92.1s] Well, I would look at where you spend your time
[94.3s] and then work backwards from there.
[96.0s] So, anything that you need to do every single day,
[100.2s] on a schedule that is repeatable,
[103.7s] and doesn't require you to be there, which is basically nine or seven of the stuff,
[108.0s] you should be automating, and you should be building custom workflows.
[111.2s] So let me show you some examples of what we build here.
[113.5s] So inside my agent OS, we have the content pipeline.
[116.5s] The reason that we did that is no one else was in loads of time
[119.3s] on keyword research, and I was using tools that I didn't really like.
[122.6s] So for example, I was using a trusted keyword research.
[125.0s] It wouldn't really give me new data,
[127.4s] and it would give me old suggestions that weren't helpful.
[130.3s] With this system, I can type for example,
[132.5s] agent OS or whatever I'm trying to focus on when it comes to SEO.
[135.4s] It will pull in that data, and it's custom made
[140.2s] to all of my websites and niche them in.
[143.5s] So if we keep scrolling down here, you can see all of the different keywords that we've got.
[148.5s] And so I'd be safe for me about 30 minutes a week.
[151.3s] Then from there, we can click on use topic,
[153.9s] and we can actually create the content and deploy it to our website and one single click.
[158.9s] That is super useful for me.
[160.4s] Now if you're watching this, you might be like, well, I don't do any SEO.
[163.3s] Like, why don't I want that?
[164.7s] And that's exactly why.
[166.0s] You want to go back into GV5.6,
[168.4s] based on what you work on day today, and tell it exactly what you do.
[172.7s] Based on your time audit and ask it to build aint.
[176.4s] So if you spend a lot of time doing research, build aint.
[179.1s] If you spend a lot of time doing lead generation,
[180.8s] add lead generation to all inside that.
[182.8s] That's how it works.
[184.0s] And this just loops around around.
[186.2s] So like, for example, when you start automating all this stuff,
[190.2s] I don't think that you're going to find you at less.
[192.0s] I think that you're going to find you work more productively,
[195.3s] and you achieve way more.
[196.4s] Like, I mean, for me, we can create way more content.
[200.8s] We can recreate more people.
[202.3s] We can grow way faster, because we have all these systems in place.
[206.0s] And so like, for me, I don't see it as like a way to,
[210.4s] three out of my time.
[211.4s] I see it's a way to save time so that I can use that to focus on other parts of the business.
[215.7s] And then just keep cycling through.
[217.6s] Okay, what do you work on?
[218.5s] What do you work on?
[219.2s] What do we need to do next and next and next?
[221.6s] Also, there's some other good really cool stuff that you can build inside here too.
[226.2s] So one thing that we actually did is, we notice,
[229.0s] okay, we have so many different agencies.
[230.8s] We've got a claw to get hermons, we're anti-gravity, we have to do a codex.
[233.8s] Why don't we add those inside a system that we can orchestrate?
[238.4s] And so we actually had a paper clip inside here as well,
[240.9s] so that we can have an organization, a team of AI agents working together.
[245.0s] They can go off and build stuff 24-7.
[247.4s] We can see everything that we've created that is really useful.
[251.2s] And also a lot of people are asking us to build this in.
[253.5s] So for example, in their proper form of community,
[255.8s] when people are asked stuff, we actually build it for them.
[258.0s] Like for example, there was someone was asking this morning about how to use chatchewity in the new app.
[264.5s] So what we did is we recorded like a one-hour course for them,
[268.2s] showing them exactly how to use chatchewity.
[270.8s] And so what I really like to do is get ideas from the community,
[275.8s] figure out the problems they have and then solve those problems with the agent OS.
[280.0s] And I think that's really good for you.
[281.5s] Like you can speed to your team, you can speed to your audience.
[284.4s] You can look at the problems that you have and solve each one of them,
[288.4s] one by one, by building the right systems in.
[291.8s] The same with the group chat here.
[293.0s] So we have a group chat and these are just ideas that were close, you can build in.
[296.6s] Because we have finished building out the experience.
transcripts/chunks/c08.mp3
BIN
View File
Binary file not shown.
transcripts/chunks/c08.txt
+59
View File
@@ -0,0 +1,59 @@
# oracle/transcripts/chunks/c08.mp3
# model: tiny | lang=en
[0.0s] sample project. So this example we actually create a group chat where you can message a room.
[6.4s] And then we can set a keyboard, Hermit, come up with a new game idea to build. It will read
[11.4s] our obsidian vaults of the memory that it tells that earlier. And then all our agents just
[15.5s] chip in and the speed to each other and the bounce ideas off each other. So like for example,
[19.8s] Joe and I are like, oh that's a clever twist. Clawd, let's do that. And then code x as I
[24.5s] know what this sounds really good to integrate over and over and over. Let's add that to our pipeline,
[29.8s] which brings me to the next setup. So we've got a paperclip for agent orchestration. We have
[34.8s] the AI agent group chat. We can also see the history of all our group chats here. Always good to add
[39.5s] history inside the agent or I suppose you won't be able to find stuff in your loop stuff and you're
[43.5s] great. Let's all substitute the confine later. And then finally we have the pipeline. So with the pipeline system,
[49.3s] you can drop an idea in here and you can go from get me idea to going into the human approval
[56.5s] section and just implement it. And this was really good because I found like a lot of people
[60.6s] including myself would have so many amazing ideas but it's very difficult to implement it for him.
[65.4s] So with this system, you can take all your ideas all to master this from the group chat because
[69.8s] the agents add this into the pipeline. And then anything that you like right, we need to build this today.
[74.8s] Okay great, let's just build it. It will come up with a plan. It will plan it out. Then we can
[79.8s] approve it. Then it will build it and then we can view what was actually built previously. And we actually
[85.0s] have a full gallery where we can see all of our ideas. And so we can go from idea to implementations so quickly.
[92.2s] Which is really really cool. So for like okay we need to build out this website, boom shout,
[96.0s] like we can open up that website here later. And again I think it's so important to have like a gallery
[101.4s] and a workspace and stuff that you can see and say for later. Because if you don't,
[106.2s] the biggest issue is that you will not be able to find it. It would just get lots of new downloads.
[111.8s] And that will go with the other 100 projects you're working on and then you won't be able to find it.
[115.7s] So that's why I'm going to a gallery and a pipeline is so so useful. Now what we also have over here is we have,
[122.3s] for example, Claude. And anytime I'm like Claude actually brings something useful out,
[126.9s] we add that feature in as well as a separate tab. I'd recommend that for you too. So for example,
[132.4s] if you saw the ultra-code update, this was amazing. You can have a 14 of agents working in parallel,
[139.3s] working together towards one task and you don't have to orchestrate it, Claude is your extraator
[143.9s] and it builds the subagents for you. So with this system we just added that as a separate tab. And it's like
[149.4s] you create your own custom version of Claude that you can edit, you can change your UI of any time you want,
[154.2s] you can come back to at any point. Then we have Hermes. So Hermes again, really, really powerful.
[162.0s] One of the things that I noticed is like they bring out new, there's always new models that we can plug into
[166.2s] Hermes and this way you want to focus on the system, not the model. So for example,
[171.4s] G556 comes out, no problem. We're going to build that out. It doesn't change how I work.
[176.5s] Doesn't make me have to rebuild anything. It's just like we can add the model inside the system.
[181.4s] The same, for example, a lot of people asking about how to have a voice activated AI agent that we
[185.8s] can talk to. And so we actually build that into this system as you can see. So what it can do is you can listen to me,
[192.2s] you can talk to me, it can build stuff out for me. And then again, we've got the full gallery here of
[197.9s] everything we've created. But we have Hermes all recall. For me, I was just find way too much time was being taken,
[205.0s] research in the latest news. And so I created this. I also think a great tip for you here if you're
[209.5s] building out your own agent or S is that you make sure it's super visual and fun to use. So for example,
[216.0s] we just added this law icon here. I think it's just a little touch. It's a little UI touch, but make it fun and easy to
[223.8s] use and very memorable. So when I see this in my archive, I know exactly what this does because I associate
[230.2s] the image with the workflow here. And if you don't like the UI of some like that, you can always customize
[235.6s] it for later too. So we have an image studio here too, but we actually have a video on a voice session as
[242.0s] well. You'll not see your previous sessions or what's been done across all of your AI agents. Which is
[246.5s] great when you see the status, we can see the skills installed into a Hermes agent, the campband, the plugins,
[252.4s] the doctor of the insights. These are all just for real cool ideas. Again, as you make it more complex,
[258.4s] more things can break. And also if you're using APIs in any part of that, just be careful because
[264.4s] you know, no more than you don't want to rinse the tokens if you're using the API XT in get expensive.
[268.5s] And number two, sometimes the API is for break or let expire and you have to be careful about too.
[274.2s] What I also like to do is let separate the outputs by the system depending on the model. And one of
[280.9s] me and by that for example is we have Hermes here. What I recommend to you is that you build out workspace
[285.9s] where you can track all your creations by the model that's created it. So when we were testing out
[291.8s] Grok 4.5 earlier this week, we have all of the creations from Grok 4.5 saved inside the workspace.
[299.4s] Really is.
transcripts/chunks/c09.mp3
BIN
View File
Binary file not shown.
transcripts/chunks/c09.txt
+97
View File
@@ -0,0 +1,97 @@
# oracle/transcripts/chunks/c09.mp3
# model: tiny | lang=en
[0.1s] We have the MCP section here, we have the actual Hermes dashboard embedded inside the system.
[5.7s] Another really good tip as well is that if you see anything open source,
[10.5s] you think, wow that would be great to build in.
[14.3s] For example, if we have a look through this list, we see what we've got here like on the room.
[19.6s] This is a free, this is a free way, a free API that you can use to build and automate anything with.
[27.3s] Now why is that important for you?
[29.3s] Well, let's say for example, using the H&S and you want to reduce your amount of tokens to use.
[33.3s] Well, you could take a project like this, you can grab the GitHub link, you can go back into chatchipity here,
[39.3s] and you can see it okay, based on this, build this into the H&OS system.
[45.3s] So, I can code for free.
[49.3s] And then that's going to take the details from OmnLyRoot, install it into your H&OS,
[55.3s] and all of the sudden, you've not just got chatchipity in Hermes in there,
[59.3s] you've now got a system with a free code.
[61.3s] And if you're wondering how does that look like in reality?
[64.3s] Here's an example.
[65.3s] So, this is a custom tool we made with OmnLyRoot, where we can code out for free,
[70.3s] we have the chart over here, we have the preview over here, we have the workspace on there.
[74.3s] Really really powerful stuff.
[76.3s] We have for example H&Y3 code, and also some of this stuff will expire.
[79.3s] So for example, H&Y3 is only available for free until the 21st.
[83.3s] And so, no problem, as soon as the 21st hits, we can delete this whole section,
[87.3s] or we can hide it, and we can do something else.
[90.3s] The same for example is to kind of figure.
[93.3s] Like, to kind of figure is an API, like it's not that nice to use,
[97.3s] and it's kind of like you have to find a place to store it, and be able to come back to it.
[101.3s] So, what do we do?
[102.3s] We build out our own custom tool, we have to kind of figure,
[105.3s] where we can see everything that's created, we can chat with it,
[108.3s] we can ask it to do stuff, and everything is stored inside one system right here.
[113.3s] And then anytime something is trending or interesting,
[116.3s] I want to play with, for example, I loop, loop engineer was something big,
[119.3s] that was, you know, everyone was talking about about two weeks ago,
[123.3s] we've loaded in the loop engineering system here.
[126.3s] And then the other thing I would say here is,
[128.3s] when you're building out agent of rent systems,
[130.3s] it doesn't just have to be for business stuff.
[133.3s] So, for example, you could have a system like this,
[136.3s] where you can generate music.
[138.3s] It could be, for example, you goals.
[140.3s] If you were really into more time, you know,
[142.3s] you can have something related to that.
[144.3s] It could be like a track of a training, it's simple like that.
[146.3s] Whatever you prefer to do, you can plug it inside the system.
[150.3s] So, you could, we have a music generator here,
[152.3s] if I want to just create some music,
[154.3s] I'm interested in for particularly for like,
[156.3s] ambium music, then I can just plug this inside the system,
[159.3s] we can get files, we can come back to the AI music we've generated later,
[163.3s] and it's already to go inside there.
[165.3s] Inside there is one.
[167.3s] So, those are all the systems that we get.
[169.3s] Another orchestration system, as well,
[171.3s] and this is with Hermes, is the Kambam board.
[174.3s] Now, with the Kambam board, what we have is the system
[177.3s] where we can drop in a task, it gets automatically triage,
[180.3s] and then all of that agents,
[182.3s] and we've got many different agent profiles,
[184.3s] get signed to that task as well.
[186.3s] So, yeah, it's unbelievable,
[188.3s] it's really changed the way that I work,
[192.3s] and the other thing I would say here is it changes all your team works as well.
[195.3s] So, if you have a team,
[197.3s] like, for me I have an SEO agency,
[200.3s] you know, we've got a team who needs to be able to use AI and that sort of thing,
[204.3s] we can give this to like our team,
[206.3s] that it is, and they can use the video agent.
[208.3s] And if we have a look down here as well,
[210.3s] these are all reviews and testimonials from the Air for Boarding,
[213.3s] you can see here, for example, like, so many people,
[215.3s] again, awesome wins and awesome results with the agent or S like Jacob,
[219.3s] he said it up in about 10 minutes and you saw that I said it up today as well.
[222.3s] And Jacob was spending like 12 hours a day trying to build out the agent OS,
[227.3s] then he just did it in 10 minutes by using our template.
[230.3s] David was saying, like,
[232.3s] Julian is cooking hard on his agent OS and anybody else seen.
[235.3s] So, it's a great way to customize it.
[237.3s] And you can see, for example, like,
[239.3s] sheena as well, sheena actually created her own version of our agent OS,
[244.3s] and used it for lead generation.
[247.3s] Now, one thing that I guess asked all the time by people was like,
[252.3s] you know, what, if you're, for example, thinking about selling AI agent services or AI automation,
[258.3s] what's the one thing the clients want to see?
[261.3s] What is in demand right now?
[263.3s] I'm for sure at the Asian Opera Insisting, like,
[266.3s] if you can build out custom workflows that solve all the problems your customer has,
[271.3s] by creating amazing systems like you've seen today,
[274.3s] so if they need help generate images, you create the image studio.
[277.3s] If they need help generate videos, you create the videos you,
[280.3s] if they need help generate social media content, you create a social media agent,
[283.3s] and how'd you do it?
[285.3s] You just go back into codex and you test it out for later.
[288.3s] So, you can easily build out all this stuff and you just grab their idea,
[293.3s] you solve all the problems you go from there.
[296.3s] Now, as well, one thing that I'll say,
transcripts/chunks/c10.mp3
BIN
View File
Binary file not shown.
transcripts/chunks/c10.txt
+51
View File
@@ -0,0 +1,51 @@
# oracle/transcripts/chunks/c10.mp3
# model: tiny | lang=en
[0.0s] He was like, you might be tempted to have Claude and Codex work on the same thing at the same time.
[7.0s] I personally wouldn't recommend that.
[9.5s] So if you are building out with an agent and you're trying to work on the agent,
[13.8s] as you're trying to improve it day by day, trying to make it better as a system,
[18.0s] the thing that I'd recommend here is that you just use one agent.
[22.0s] So for me, for example, I'm using Claude right now and I use Faber 5 to help me,
[26.0s] you might prefer using chatchipy to you might prefer using Hermits, but seek to one.
[30.0s] Because if you have multiple agents, it multiple different apps, number one that gets really messy when you're building out an agent offer in system.
[38.0s] And number two, they could like overwrite or delete something that the other agent is building.
[45.0s] So for example, Claude is building out something that's just broken.
[48.0s] Chatchipy might be like, let's remove that from the code.
[51.0s] And then all of a sudden you've got two agents kind of interrupts any chance work and that can get super messy.
[56.0s] Wouldn't recommend that.
[58.0s] One thing that I haven't tested, that would be very interesting in testing mode, is with Codex is very easy to add plugins.
[66.0s] Now, what would be very interesting here is if you could add those plugins inside the agent or S as well.
[73.0s] So that's an example that we've got email and Gmail plugged into Codex, right, into Chatchipyty work.
[80.0s] So you can send the email address, you can manage your email.
[83.0s] So you can draft replies for us and if it else.
[85.0s] Now, what would be very interesting is say to Codex, a build out this scheduled task into my agent or S, so that we have the email agent inside there.
[99.0s] I'm not going to do that right now, because I think it's a task that we'll probably take about 20 or 30 minutes to really perfect.
[104.0s] But that would be interesting to take for the plugin section from Codex, and start adding that into your agent or S, so that you can automate and build anything as you want right there.
[115.0s] Other things that I would recommend.
[118.0s] Number two is you might be tempted to put this on a website, for example, like a lot of our websites we use with NetLefite.
[127.0s] I wouldn't do that just because again, like I think you need to keep it secure in everything else.
[132.0s] Also, if you want to reduce the amount of tokens, you can also use stuff like Free APIs and you can use local models.
[140.0s] So if we go to the local section here, we can use local models from Olammer or LM Studio, and this is particularly good for like the tasks that a quite time consuming, but don't need frontier levels of effort.
[155.0s] So let's say for example, we're looking at all of the models on Olammer.
[161.0s] I'm like, okay, or NetLef, or NetLef looks awesome, right?
[165.0s] Actually, it's a really good tool. I've test it out.
[167.0s] Persona, we have the local lead board on Goldiebench.
[171.0s] We've all the local models, and you can see what we've created in T.R. web, Formula Finals.
[175.0s] And so you might be tempted to build that into your agent or S as well.
[179.0s] And I think it's actually pretty interesting to build out a local code.
[183.0s] And so what we have over here is we have this local section, and we have the preview of F and created what the works face of our previous generations, and we can use local models in SOT System 2.
[195.0s] And that's another way to use it for free, because then you are a combination of like Omniru, which is free way to use like 90 models, and it kind of orchestrates across them.
[204.0s] You have HY3, which is another free API, you can plug into the system, and then you have local models, which are basically free forever once you've installed them.
[212.0s] So between those two things, those free things, you can really reduce the amount of tokens the years as well.
[219.0s] What else can we do here? The other thing that I like to do is when I'm testing out all of the models, with the same agent, the builds out the HNOS, with all the creations I just put them on Goldiebench so that I can document the process here.
[234.0s] There's another good way to generate leads, because people are searching for interesting stuff, people aren't in no what you've built with it.
[239.0s] People are interested in your agent or SOT if you actually share what you're working on and everything else, especially if you're working on some new unique stuff.
[246.0s] So you can actually give your agent or S access to NetFies well, and then it can document the whole process, the whole building of the agent or S on what you've done.
[256.0s] Kind of like a journal, but it's not you update, and it's not you writing this agent, which is really cool to us.
[262.0s] So that's basically it. That is the whole system. How to build an agent or for an assistant?
[267.0s] How to customize it as much as you want? How to add a memory inside there? Which agent do you use to build it?
[272.0s] How to reduce the amount of tokens you've won. Also, how to create a system that actually works rather than just breaks?
[280.0s] Also, how to use it in a simplified way using our templates. Now if you want to get my full agent or for an assistant, you can get that inside there,
[287.0s] or for boarding inside the Aginal S system section right here. So we update the stately and get the set file for it and you can get the last, you see the last update date here.
[295.0s] You can get a full guide and how to use it, and then every time we build something new.
transcripts/chunks/c11.mp3
BIN
View File
Binary file not shown.
transcripts/chunks/c11.txt
+35
View File
@@ -0,0 +1,35 @@
# oracle/transcripts/chunks/c11.mp3
# model: tiny | lang=en
[0.0s] is full. So for example, like we create a chat GPT 5.6.1 out course, you can get that over here as well.
[6.0s] And then also if you want more help and support on building out with the H&O S, you can post inside the community
[12.0s] and we reply to you and create video tutorial for you ever single day. So for example, like Matt was asking about Ome and the best way to use it and what you can use instead,
[20.0s] we solve that problem for him and we reply to everyone and just help each other as much as you can in the best way that it can as well.
[27.0s] Inside the classroom you get access to all of our best trainings. If you're a complete beginner, you can go from begins to expert inside this section.
[34.0s] And inside the calendar, you can jump off weekly coaching course. Now this is super useful because you can share your screen.
[40.0s] You can ask questions you can get help and support him. And you can also meet other community members who are building out this sort of stuff as well.
[48.0s] Inside the map is well, you can meet people locally who are building out with AI agents like you and the soul available inside the AI Prof.
[56.0s] or link in the comments description or just go to the AI Profible.com.
[60.0s] One other thing that has been really useful when we're building out all this stuff is that when a model, like for example, doesn't work the same way or you can't use it anymore.
[72.0s] So for example, if a great example now, because you've got a system that improves everything with day, it doesn't really matter about that sort of stuff.
[80.0s] You don't need to worry about that. So if GVT5.6 comes out, that's great. We can add in.
[85.0s] But if, for example, Fable5, I've got taken down, which you did a few weeks ago, it doesn't matter because in like two minutes you can go back into chatchipity and you can say,
[96.0s] you can make remove the integration with Fable5, add this in, etc.
[101.0s] And also one thing that's really useful is because you've named the conversations, you can come back in like two weeks and edit that conversation.
[109.0s] So for example, what I like to do is I applaud if I'm scrolling down. If I need to improve the setup that we have with Pony's now, which is another AI project, I can go back inside the chat here.
[120.0s] And I can say, hey, add this, change that. And because everything is named, it's super easy to organize and it's easy to see everyone as well.
[128.0s] So my very interesting thing is that we've set up two in the past is like computers with the agent OS. So I can go inside her, for example.
[136.0s] And I can say, hey, Hermes, can you just open up Google for me and that sort of thing?
[142.0s] You can actually do that with Apollo, which is the voice agent and that actually has access to computers as well. So it can edit local files, it can edit.
[150.0s] You can open stuff inside your browser and everything else as well. Which is really, really good.
[155.0s] And that's pretty much it when it comes to the whole system. I mean, it's so much fun to build. It's so inspiring to build. It's also great to just build something that feels like your own.
[166.0s] It's never been more fun to use than I think this is only been possible this year. Like, if you went back last year, it was great using AI, but you couldn't build in the same way, but models would just not add this level.
[179.0s] Whereas when you're building the agent offerings, it's now any custom workflow that you have, things that you've never even dreamed of creating before, you can just add inside this, you know, you can plug in like, right, when I just want to delete that, no problem at all.
[194.0s] So it's so much fun, so much fun. So more things that you can do as well is you can get skills and then you can plug them into the system.
[203.0s] So for example, like there's a skill for reducing tokens, which is called Caveman, and with Caveman, it basically reduces the amount of output tokens you plug in.
[212.0s] And you could add that inside the system as well. The final thing that would say, and this is a little feature is that you can delete stuff, give yourself the ability to delete stuff.
[221.0s] So you can go into chat, cheapety, and you just say, okay, make sure that everything that we create inside the agent offerings system, we can delete so that I can remove anything that doesn't work.
[230.0s] Because like, sometimes you might generate something with a gelinfoam to, but then it actually doesn't work very well.
[236.0s] And so how do you get round that? Well, you can just make sure that you have the ability to delete that and just add a delete icon for later.
[244.0s] So that is basically it. That is a full one hour course in how to build agent offerings system, how to automate anything, how to build your own custom workflows. If you want my full setup, you can get that inside the air, pop water and you can get the amazing community that's all using this sort of stuff.
[259.0s] You can connect me personally, you can direct messry anytime you need anything. And thanks so much for watching. I'll see you in the next one. Cheers, Baba.
transcripts/ep_build_agent_os.clean.txt
+105
View File
@@ -0,0 +1,105 @@
# How to Build Agent OS - FREE Course! — AI News Today | Julian Goldie Podcast
# Transcript (cleaned) — ~59 min episode
# Source: https://open.spotify.com/episode/4HWHi9beiAW8JGi0CFVgTV
# Resolved RSS: https://feeds.transistor.fm/ai-news-today-julian-goldie-podcast
# MP3: https://media.transistor.fm/618f447d/afb640ea.mp3
# ASR: faster-whisper tiny/int8/beam1 (CPU), transcribed in 12 × 5-min chunks.
# Proper nouns restored. UNCERTAIN terms flagged inline as [??].
INTRO
This is a full one-hour course on how to build your own agent operating system, based on what's working for me. An "agent OS" is a way to have all your agents inside one dashboard, where you build custom workflows and have everything one click away.
EXAMPLE OF WHAT'S POSSIBLE
- A video agent that creates videos in one click.
- An SEO content pipeline: research keywords, generate content, deploy to the website in one click.
- Hermes Jarvis — a voice-operated AI agent that looks and feels how I want.
- Hermes Oracle — pulls trending news.
- Hermes Astros — a custom workflow that takes latest news from Twitter/X, gives trending topic ideas, suggests titles, then plugs into the video agent to automate video; also feeds NotebookLM to create free infographics.
- A lead-gen + email outreach tool: type the leads you want, it generates a list, writes the email campaign, sends it, pulls in lead details, and manages metrics (validated / reachable / sent / replied). Fully automated inbox — one beautiful dashboard.
He notes you can build all this free using your existing subscriptions (e.g. ChatGPT, Claude, Codex — already paid). He used GPT-5.6 [?? "G55.6"] as the brain and "Fable 5" [?? real product?] to help build.
STEP 1 — BUILD THE DASHBOARD
- He uses ChatGPT to build a "Mission Control" dashboard to operate ChatGPT, Codex, and Hermes agent together. Prompt: "create a mission control dashboard where I can operate ChatGPT, Codex plus Hermes agent. Make it clean, beautiful, interesting to use, make sure it actually works."
- You can attach a screenshot of a style you like for inspiration.
- Two install paths: (a) his paid template (updated every time he codes, shipped as a setup/zip file), or (b) build from scratch inside ChatGPT.
STEP 2 — INSTALL & CONFIGURE
- Run it locally on a separate local port. He recommends local over VPS for security: if you give a cloud agent file-edit + computer control and expose it to the world, it's not secure. VPS is fine if you limit the tools.
- "Fast mode" toggle for faster outputs.
- Two instances: template version + from-scratch version. Avoid port overlap (one on port 3200, one on 1200).
- Let it configure parts that don't need an API.
STEP 3 — MEMORY SYSTEM (Obsidian + Omi)
- "Focus on the system, not the model." Hide/replace tools as the landscape changes (he mentions OpenRouter [?? "OpenCore"] was huge two months ago, now less so; GPT-5.6 mind-blowing, but GPT-5.5 not usable for this).
- Obsidian is his preferred memory system — free, open source, stores everything in markdown files in an organized system. Agents organize it.
- Memory galaxy metaphor: every dot = a memory (person/project/business/tool), all linked. Agents understand what you've worked on → better, more personal responses.
- Omi [?? "omie"]: free open-source screen/mic recorder (MIT license, basic tier free). Exports notes directly into Obsidian. "OME → Obsidian."
- You don't build the memory yourself — tell Claude/agents to create and organize Obsidian files.
- Keep the old agent OS running on its port while building the new one (don't let it overwrite).
KEY PRINCIPLE — SIMPLICITY
- The simpler you make it, the better. You won't use every feature. Hide tools you don't use (e.g. he hasn't used "Kimmy code" [?? Claude Code?] in weeks).
- Build only what's relevant to you.
TESTING & ITERATION
- Test it: "are you working?" — he found the from-scratch build created a nice UI but didn't actually test/reply (blank screen). Template version worked immediately.
- Use ChatGPT (or Codex, or Hermes) to build. ChatGPT is more visual. You can fork conversations — run multiple agents in parallel inside the same project folder (e.g. video agent + SEO agent + lead-gen coded side by side).
- Plug in existing subscriptions to avoid API token cost: Claude CLI, Codex (already subscribed) → build and use the Agent OS for free. "You can build these agent cooperative systems for free."
MEMORY INSTALL INTO AGENT OS
- Tell the agent OS: "add the memory system from Obsidian... install it... make sure ChatGPT and Hermes are synced to my memory system. When I use them, update Obsidian; when I ask questions, learn from the Obsidian vault as context." Everything personalized.
- Three layers added in ~27 minutes: front-end, custom workflows, memory.
IMAGE GENERATOR EXAMPLE
- "Create an image of a cat eating biscuits" → Codex figures out aspect ratio (cinematic) on its own. He prefers letting the frontier model make decisions — it has more attention/time than he does.
- Found a real bug (500 error) during testing — valuable. Renaming conversations keeps track of what you're building.
- Token usage warning: fast mode + parallel building burns tokens fast. He hit the usage limit, used weekly reset. Dial down effort/speed/model (e.g. to "4o" [?? "25.5"]) for basic tasks.
TOKEN MINIMIZATION
- 10 strategies in his "token minimization playbook" — claims up to 95% token savings / 2-3× output.
- Templates save tokens (10-20 min vs hours from scratch).
WHAT WORKFLOWS TO ADD
- Do a time audit. Automate anything repeatable, scheduled, that doesn't need you present.
- His content pipeline: keyword research → create → deploy to site in one click (saves ~30 min/week).
- "Based on your time audit, tell GPT-5.6 what you do and have it build custom workflows."
AGENT ORCHESTRATION
- "Paperclip" [?? agent orchestration tool] — a team of AI agents working together 24-7 (Claude, Hermes, Antigravity, Codex).
- Group chat: message a room, agents chip in and bounce ideas off each other, reading the Obsidian vault as shared memory.
- Pipeline system: drop an idea → human approval section → plan → approve → build → gallery of what was built. Idea-to-implementation fast.
- Gallery + workspace: critical so you can find past builds (otherwise lost among 100 projects).
- Claude "ultra-code" [?? subagent feature]: 14 agents working in parallel, Claude as orchestrator building subagents.
- Voice agent (Apollo): listen/talk/build. Full gallery of sessions across all agents.
- "Make it super visual and fun" — small UI touches (icons) make workflows memorable.
MCP & OPEN SOURCE PLUG-INS
- Embed the actual Hermes dashboard inside the system.
- If you see open-source tools, grab the GitHub link, tell ChatGPT "build this into the Agent OS" → code for free. Example: "OmnLyRoot" [= OpenRouter] for free coding; "HY3" [?? a free OpenRouter model] available free until the 21st, then hide/delete it.
- Custom tools: e.g. his own "Figure" [?? a figure/diagram tool] wrapper storing everything in one system.
- Non-business uses: music generator (ambient), goals tracker, etc.
- Kanban board: drop a task → auto-triage → assigned to agent profiles. Changed how his team (SEO agency) works.
TESTIMONIALS / SELLING
- Jacob: set up in 10 min via template vs 12 hrs/day building manually.
- Sheena: built her own version for lead gen.
- For selling AI services: build custom workflows that solve each client problem (image studio, video agent, social media agent).
BUILDING TIPS (what NOT to do)
- Don't run Claude AND Codex on the same thing simultaneously — they can overwrite/delete each other's work. Pick ONE agent to build/improve the OS (he uses Claude + "Faber 5" [??]).
- Don't put it on a public website (Netlify) — keep it secure/local.
- Use free APIs + local models (Ollama, LM Studio) for time-consuming, low-effort tasks. "HY3" free API + local models = near-free operation.
- Document builds on Goldiebench [his benchmark] — also a lead-gen channel.
- Give Agent OS access to Netlify to auto-document the build journal.
CLOSING / PITCH
- System survives model churn: "if Fable 5 [??] gets taken down, in two minutes you remove the integration and add another." Named conversations make it easy to revisit and edit weeks later.
- Voice agent (Apollo) has computer access — can open Google, edit local files, open browser.
- Skills can be plugged in (e.g. "Caveman" [?? a token-reduction skill] reduces output tokens).
- Always give yourself a delete ability for anything that doesn't work.
- Full setup + community (AI Profit Boardroom) inside the Agent OS section: setup/zip file, full guide, daily video tutorials, weekly coaching calls, local member map. "AIProfit.com" [??].
---
NOTE: This 59-min episode had heavy ASR corruption. Confident restorations: ChatGPT, GPT-5.6, Hermes, Agent OS, Obsidian, Omi, Claude, Codex, Antigravity, OpenRouter, Ollama, Netlify, Apollo, NotebookLM. UNCERTAIN (flagged [?]): "Fable 5"/"Faber 5" (a real product he references?), "OpenCore"→OpenRouter?, "HY3"/"25.5" (model slugs), "Caveman"/"Kimmy code" (tool/skill names), "loop engineer" (concept), "Pony's now" (a project name), "Rue sub-brain" (likely "free sub-models"). Promo URLs ("AIProfit.com", community name) were unintelligible in audio.
Raw verbatim: ep_build_agent_os.raw.txt (same dir). Chunks: chunks/c00.txt..c11.txt.
transcripts/ep_build_agent_os.mp3
BIN
View File
Binary file not shown.
transcripts/ep_build_agent_os.raw.txt
+863
View File
@@ -0,0 +1,863 @@
# oracle/transcripts/chunks/c00.mp3
# model: tiny | lang=en
[0.0s] A gented offering systems full one hour course,
[3.5s] how to build an automate anything.
[5.8s] So today I'm going to show you exactly how to build your own agent
[8.9s] offering a system based on what's working for me.
[11.0s] If you're wondering what an agent offering a system is,
[14.0s] basically it's a way to have all of your agents
[17.5s] inside one beautiful dashboard,
[20.2s] where you also build out custom workflows
[22.9s] and you have everything one click away.
[25.1s] And the amazing thing about this is
[27.4s] that you can basically have everything in one single place
[31.6s] that's ready to automate and offer it as you want.
[35.0s] So let me give an example.
[36.6s] For me, I suspend a lot of time create videos,
[39.3s] and so we create this video agent, the basically does it for us
[42.5s] in one single click.
[44.0s] We have, for example, an SEO content pipeline where we can research stuff,
[47.9s] we can find keywords who can generate content,
[50.2s] and then we just deploy to our website in one single click.
[53.3s] And basically everything that you want to automate,
[55.4s] you can automate using these systems.
[57.4s] So for example, Hermes agent, not very fine in the thermal.
[60.7s] If you're just using the thermal, not that great.
[62.7s] Even if you're using for example Hermes desktop,
[65.1s] you still can't get it exactly how you want.
[67.9s] With this system, for example, we have Hermes Jarvis, which is a voice-operated AI agent
[73.3s] and looks and feels exactly how I want it to look.
[75.7s] It's got all of my projects over here.
[78.1s] And what else I have, for example, at Hermes, I'll recall the pause
[80.5s] and let's trend it news and if it melts.
[82.1s] And we're not seeing that to both stores to show off my own own product.
[84.9s] I just want to show you what's possible.
[86.6s] With this sort of stuff, for example, this one,
[89.0s] this is Hermes Astros, which again, is a custom workflow.
[92.0s] And you can rename this, you can change it how you want,
[94.9s] you can build whatever you want using these systems.
[97.4s] And this, for example, actually takes the latest news from Twitter,
[101.2s] gives us the trending topic ideas.
[104.9s] Looks at different titles we could use for that content.
[108.3s] And then also we can just plug out to the video agent.
[110.8s] So it can automate the video.
[112.0s] We can plug it into notebook, lemon.
[113.5s] Create, for example, free infographics and all that sort of thing.
[116.1s] And then we could create a serial content for you.
[117.8s] And you can build all of this for free using your existing subscriptions as well,
[122.3s] which is very important to note.
[123.6s] So, for example, if you have chat cheapity already,
[126.4s] and you already subscribed to it,
[127.9s] well, you can actually use chat cheapity to build all this out.
[131.4s] I actually personally used Opus 418 inside,
[134.4s] call it an Avise Fable 5 to help me a lot along the way.
[137.0s] And for example, if you have something that's a problem you want to solve,
[141.4s] let me give an example.
[142.6s] So a lot of you ask me like, how do you automate leads and email outreach using something like
[147.1s] Hermes?
[147.8s] So with Hermes, we have a tool here.
[149.6s] The congenerate leads, we just type in exactly what type of leads you want to get here.
[154.2s] It generates a list leads.
[155.8s] Then it has all the campaigns.
[157.2s] It can write the email campaign itself.
[159.3s] Then it can actually send the emails with these campaigns,
[161.9s] it pulls in all the lead details.
[163.7s] And then from there, we can manage the metrics.
[165.7s] We can see how many leads we've got, how many we've validated,
[168.2s] how many ascendable, how many we actually send, how many replied,
[171.1s] and we've got our inbox and sent inbox fully automated with Hermes agent,
[175.8s] it's like one beautiful dashboard.
[177.4s] And that just saves so much time.
[179.7s] So this is the power of this stuff.
[181.4s] This is how amazing it is.
[184.0s] And you will learn how to build all this today.
[187.0s] And I'm going to show you exactly how we can get started.
[190.4s] So if you're learning why I'm the right person to show you this,
[195.3s] I literally spent about three to four hours a day
[198.7s] coding this out and making it better and better.
[200.5s] And there's so many mistakes that I learned along the way,
[202.4s] so many things that I can prove.
[203.8s] So with that, let's get straight into this.
[206.8s] So I'm just going to be using G55.6 as an example.
[209.6s] I quite like soul.
[210.6s] I think it's great to build with.
[212.2s] And you can see that we've got GVT and chatGVT work over here.
[216.5s] It's sort of snap.
[217.7s] Now for example, we set up Hermes agent.
[220.6s] That's fully ready to go with use GVT.5.6 as a brain.
[225.3s] Inside Hermes agent and now we want to say create a dashboard with it.
[229.8s] So what we can do here is we can actually go inside the chat here and we can say, okay,
[235.6s] now create a mission control dashboard where I can operate chatGVT,
[244.6s] codex plus Hermes agent.
[249.5s] And we can say make it clean, make it beautiful, make it interested to use
[255.4s] make sure it actually works.
[257.0s] And then if you have an idea of what you want to create,
[259.8s] so let's say for example, we have a certain design style with that sort of thing,
[262.7s] we can actually use that for inspiration when you're building this out.
[266.6s] So if we wanted to build this setup inside chatGVT,
[270.6s] we can just take screenshot, plug in and just say,
[274.1s] I've attached an example of the style I wanted to create.
[278.5s] And so we can hit enter now and that'll begin building this out.
[283.4s] So if you want to start from zero and you want to build it exactly how you would like it,
[289.5s] in a way that's easy to use, you can go from there.
[292.4s] That'll also show you another way that you can do this.
[294.6s] So if you want to use someone else's template,
[297.8s] so for example, we have a
# oracle/transcripts/chunks/c01.mp3
# model: tiny | lang=en
[0.0s] version of mine inside there, but bottom that we actually update every time I code out for
[4.8s] three or four hours, I will actually save that. So you can see the sit file here. If you
[9.8s] want to do it, how do you install this? See we do the same thing. You can use whatever agent
[15.3s] you prefer so that could be checked with the it could be for example called whatever you prefer.
[20.3s] And we can start a new project. Start from scratch, save that new project. And then we can
[26.7s] add this so we can add the sit file. So we can grab the sit file over here, which is
[34.7s] there. And we can say install and configure this agent or S on a local port we don't already
[43.4s] have running. So I like to run these things locally as well. And we'll come onto that and
[48.2s] why in a second. But now we have two options. So we can use the setup from the Air Profit
[53.7s] for the bottom and that's just kind of like ready to go and all the other option that
[57.6s] we have is if you just want to build this out free from scratch you can go and build it inside
[62.9s] chat to you with you as you can see. So now it's going to say okay what should the dashboard be
[68.1s] able to do. So says before implementation, I need three decisions that material effect the
[74.8s] architecture. So we have Asian control. We have workspace. We have application format. Now for
[81.1s] me, I'm going to say just allow it to run locally for me in Chrome. And have one fix to
[90.4s] folder and allow it to be fully operational. So this is something important to note. CC when it's
[96.2s] asking you it's asking you like do you want it for operations so that it can actually like edit
[99.6s] files. It can do things autonomously for you or do you want it kind of like a chat. For me, I
[104.8s] don't want a chat because if I just want a chat I'll just go to chat to chat to be here.
[109.8s] And that is exactly why I actually created the agent OS because when you're just going back and
[116.1s] forth inside the chat here, it's super limited. Like it doesn't feel authentic, it doesn't feel
[120.1s] powerful. It doesn't really have everything you want and it's very, very limited in terms of what
[124.8s] it could do. So that's why I prefer to be like way more powerful. The whole point of this is to build
[128.9s] something absolutely amazing that can be a little bit of an elephant and just runs like magic. So that's
[134.2s] why we are setting up like this system. So you can see over here this is beginning to install now as well.
[142.9s] Alright so you can see that we've got two options here. We have the agent OS been installed locally from
[147.6s] template and we have the Hermes agent OS over here too. See two different options. Now both of these
[152.8s] will run like you can see I actually want to switch this to fast mode. So I'm going to say switch to fast mode now.
[159.1s] Just so that we can get faster outputs from both of these. That's already enabled which is great.
[164.4s] And now when we're using here you can see that it's beginning to use this local setup. So you'll see
[172.0s] a number like this. Now you might be wondering why you're running it locally, why not just run it on VPS.
[177.5s] Here's something to consider. Okay and this is something that I saw when I was looking at all in the
[182.5s] stakes from open core and a lot of the issues that people face is that there were two things with agent. So you
[188.2s] have to be very careful of. Number one is that you have to be careful because when you're building
[194.3s] with this sort of stuff, if you give access to multiple different devices then someone accidentally could
[202.2s] access your agent OS and if it can edit files if it can control your computer if you can do everything like that
[207.9s] it's not that secure. If you if you've given the world access to it. So that's why I don't recommend that. Instead
[213.9s] what I recommend is that you use the system like this and you run it locally. Now you can run it on
[219.4s] VPS. We have plenty of AR profitable and members who do that and then they just limit the tools that
[224.2s] they give access to but it's up to you how you want to do it. Now you can see that we already have the agent
[228.9s] OS installed over here. So this is fully installed as you can see here. We've got a friend setup. It's set up on a
[236.0s] separate agent OS local port. So for example if we go over here this is my actual agent OS and then we have
[243.6s] the template over here. So it's pretty easy and the thing to note here is when you're using this you just need to
[248.8s] configure everything. So it's now running using the template that gave it but I'm going to say configure it
[255.7s] here. I'm just going to make this simple if you're doing the sound. We're going to rename this project and we're
[260.8s] going to call that template setup. And then we're going to use this project here and we're going to call
[268.8s] starting from scratch. We'll just call this new agent OS example to a two different options right there.
[279.6s] So now it is said okay before I configure it just we're just going to go through step by step.
[288.9s] Set it up at each part that doesn't require an API and we're going to say yes and figure it up.
[294.7s] Right just so that it can start working directly now we have
# oracle/transcripts/chunks/c02.mp3
# model: tiny | lang=en
[0.5s] the other setup here. So this is the upper version working locally. As you can see and then I just want to make sure one thing as well
[8.7s] that we don't have overlap in ports. Yeah, so this is
[12.1s] starting on this port which is 3,200 and this one is running on that port which is 1,200. Now you actually see when you're doing this like it does a very good job of
[24.4s] building
[25.4s] how locally and also emulating the style that we want. So we'll see how this goes.
[34.0s] So it says course, scaffolding is in. I'm wearing this out. Now while we went for those two tasks both the new one and the template version to be done,
[44.5s] something else to consider is the memory setup. Now there's so many different memory options.
[50.2s] For example, if you've never come across it, Hermes agent memory, so many different options for that. You can use loads of different assistant memory. There's lots of different options for this.
[61.2s] The reason that you would have a memory is so that you're not starting from scratch and also that you'd link everything together.
[67.2s] So if we have a look for example at this setup over here. When you're building this out, all of these agents can use the context from our memory system. Why would we want that?
[80.2s] We would want that because for example, one of the reasons you want to have an agent or S is that you don't go in between different apps and have to configure everything separately.
[88.2s] So normally what happens is this is a big problem for most people is like they will go over to chat,
[95.2s] to chat, go over to chat, and then go over to the cloud and these two things they don't think together. They don't work together. So what you want it said is a memory system. How do you build a memory system? You can go into Obsidium Obsidium is my preferred memory system.
[110.2s] I'm sitting in basically stores everything in my town files in a beautifully organized system like this.
[116.2s] Now if you're wondering how to get Obsidium, you can just go to Obsidium and it's a free app that you can download. It's actually open source.
[124.2s] So you can just get Obsidium and then you configure it like that. It won't look as beautiful like this, but you can ask the agents to organize and smart and help you and organize it properly so it looks more efficient.
[138.2s] So step one is building out the dashboard which we've got chatchipity working on over here. Then step number two is that we are building the memory system inside all of this.
[150.2s] So that when we're using cloud and when we're using chatchipity and if we're using hermys they all sink together. They all work together.
[159.2s] So the reason for that is that then the agent is no exactly what you've worked on which means you'll get better responses. You'll get more personal responses and it'll be easier to use each one.
[170.2s] So if we have a look at our memory system over here and we scroll in this is our memory galaxy. So every single dot that you can see is a individual memory and this is all linked together inside one beautiful galaxy.
[184.2s] Where we've got all of our files over here that we can preview and we can have a look at really easy to use.
[191.2s] So when we're doing this it means that we can sink our memory together and we can have every agent basically understanding what the operations have worked on.
[202.2s] And the other thing to note about all of this system is like everything all dot inside here is for example like a person or a project or a business or a tool that I've used and it all gets sinked inside one place.
[215.2s] Now when you're doing this swap you don't have to create all of this yourself you should get your agents to do it. So you can say to for example, Claude, hey Claude, based on everything I've worked on recently just create an obsidian file and organize that and then you can say to chatchipity.
[231.2s] Based on my local obsidian file can you just improve this but that's how it works.
[236.2s] And one thing I'm just going to do here is I'm going to say don't remove the previous system that was set up on the local port just create a new one.
[247.2s] So we still need the agent or else on the old port because what it's done is basically take a mile port out and then replace it.
[254.2s] So make sure the old one on that's famous this local port is back, it's back up ASAP and then you can use it totally different local host.
[266.2s] So that's something to note as well as like when you're running this locally if you're running in other projects like other get her projects in the background.
[273.2s] You don't want to have them conflicting because you can only run one local project on one local year.
[282.2s] So we've talked about memory and we've talked about how to do dashboard now also it's important to note with the memory as well something that'll make it a lot easier.
[291.2s] You can use something like omie and omie can actually record your screen take notes can listen to your microphone.
# oracle/transcripts/chunks/c03.mp3
# model: tiny | lang=en
[0.0s] phone, you can understand what you're working on and then it can create a list of memories
[3.7s] here.
[4.7s] So you can see for example, 21 minutes ago, one hour ago, one hour ago, it understands
[8.8s] everything that will work on recently.
[11.4s] That's really useful because then I don't have to stay at the nose myself and actually
[14.6s] get more detail than any of the other things that I'm working with.
[18.1s] So when we're using OME, which again is another free open source project, if you think
[23.6s] they do have a subscription if you're using it a lot, but if you're just using a basic version
[28.8s] of it, you can get access right here.
[30.8s] It's available on OME, it's MIT license, it blisters for you, understands what you're
[36.9s] working on, takes notes, and then you can actually export that to Obsidian.
[41.7s] So what OME, that goes to Obsidian and we go from there.
[46.0s] So if we go to Home inside OME and then we go to the settings and we check this out, we can then export
[57.8s] this directly to Obsidian as well, which is really, really useful.
[63.1s] So this is being an out now.
[65.3s] You'll see here, this is quite a big project for chat activities to work on.
[72.4s] So when it's being an out from scratch, you can see here that's beginning to do it out, said,
[76.2s] I said, I'm working on FIM, prioritizing the UI and the integrations here.
[85.2s] But it might take a few minutes, it's been coding out for like 10 minutes already, just working
[89.2s] on this brand new H&OS.
[91.0s] Whereas for example, as you saw before, we've already got the H&OS live with the templates, so it's
[97.4s] way faster to use the template than it is to use a brand new set of from scratch.
[103.8s] So if you want to save time, I would just go with the template inside the app, I'll put it in,
[107.3s] but if you want to build your own from scratch, this is how you can do it over here.
[111.0s] Now when we're waiting for those to run, one other thing to note here is that once we're
[116.8s] building this and we're creating stuff out, you only need to add what's relevant.
[122.8s] One feature that I built in and I recommend this for you too, is that you have an option to hide
[127.9s] things, because what we're trying to build here is something where we focus on the system,
[133.1s] not the model.
[134.1s] So for example, OpenCore, go back like two months ago, this was probably the biggest A-A-A-Gene
[140.5s] in the world.
[141.5s] It's only recently that people stop using it, and that includes me because it's Hermes to
[144.7s] go over.
[145.7s] So my point here is that you want to focus on the model, on the system, not the model.
[153.3s] So you can hide OpenCore and you can add it in.
[157.2s] And I would allow that for yourself as well, so that you can move things around, you can change
[161.2s] it, because things in AI are crazy.
[164.0s] Like G556, as soon as it came out, absolutely mind blowing.
[168.2s] But G555, I would never build anything like this with it.
[172.0s] So as you go along here, there's going to be things that you swap out that you don't need.
[177.0s] I mean, if you look at my setup as well, because I'm testing some of the things all the time,
[181.7s] I have loads of tools here, but for you personally, you might want to use Hermes or you might
[186.1s] only use Clawd and you just need to build out several workflows of each one.
[189.8s] So that's something important to note, it's like you don't have to use everything that you
[193.9s] see here.
[194.9s] It doesn't have to be as complex as this.
[196.8s] One thing that I learned along the way was the simpler you can make this the better, because
[203.2s] you won't use every feature.
[204.2s] For example, Kimmy code, I personally haven't used for weeks, so I should probably hide them.
[210.3s] I can always reuse it if Kimmy K-Table and A comes out and then it's absolutely mind blowing.
[216.2s] So you see how you can swap things in, you can swap things out, etc.
[220.1s] You've got the mission control here.
[223.1s] The daily goals list as well.
[225.2s] This is really useful.
[226.2s] I'd recommend building this into.
[227.6s] In any feature that you see here, that you like right, I want to add it in, or for example, let's
[232.0s] say you have an idea and you like I need to add that in, I need to add this in, etc.
[235.8s] Well, when you're building any of this out, you can go back into chatchipathy and you can say,
[240.6s] okay, great.
[241.9s] Now build out a daily journal section or now build out a daily habit tracker.
[248.5s] What are the features you have in your mind?
[251.2s] You can build in.
[253.2s] And you can ask chatchipathy to build in.
[255.4s] And if it's taken ages to reply, you can just say, okay, add a journal feature.
[262.2s] And if you type forward slash steer inside or if you add the message here, it will actually give
[269.1s] you the option to steer it in the right direction if it's in the middle of coding.
[273.7s] So that's how it works.
[274.7s] I said by said, now if you have a look, we've got the mission control template installed.
[281.2s] So this is actually the template from the AAPRF boarder.
[285.4s] If we have a look at the setup from the agenda, go S, the chatchipathy is built.
[290.8s] It looks similar, but this is the version where you're signed from scratch.
[295.6s] Now when you build an inside set by set, actually looks really nice.
[299.6s] When you build an inside set by set, actually looks really nice.
# oracle/transcripts/chunks/c04.mp3
# model: tiny | lang=en
[0.0s] I said by said, you want to test it out.
[2.6s] So let's just test, are you working?
[5.2s] And you'll see things that you want to improve.
[6.6s] You'll see things that you want to make better.
[8.2s] And actually, if you have a look at this, it totally blanks.
[11.0s] So we said, are you working?
[12.9s] And then it just didn't reply.
[14.6s] So when we fix things to the flight, this and this way,
[17.5s] it can take quite a few hours per day
[19.6s] for me to test out and improve this.
[21.7s] We need to drag that screenshot in and say, hey,
[23.9s] this isn't working.
[25.3s] Can you fix it?
[27.4s] And we'll delete that previous response of the gig.
[30.0s] For example, is not replying and also when I click enter,
[34.4s] actually just had a blank screen.
[36.1s] So just make sure you test and actually make sure it works.
[38.6s] So it's much easier to start from a template,
[41.5s] a template, why actually work, whereas, for example, this set up one.
[44.6s] And you see how this is running through each stage of the process,
[48.8s] whereas, for example, with the new version from scratch,
[52.1s] it created a nice UI, but it didn't even test it,
[55.0s] even though we previously asked it to.
[56.4s] And that's on 2.5.6, so extra high.
[59.1s] So because it's such a big project and because it was coding for so long,
[62.2s] you can't have forgotten to do what we asked it to do.
[64.5s] Because we said, make sure it works.
[67.1s] Make sure it actually, you know, actually works,
[70.6s] set by set, but it doesn't, it just created the UI.
[73.0s] So with what the UI, the user interface that's looking good,
[77.0s] we just need to make sure that the rest of it is working to.
[81.0s] So it's going to start improving that as well.
[83.4s] So if we come back to scratch, you'd be fine.
[85.5s] That's beginning to analyze it as well.
[87.8s] Now also the great thing about chatchipathy,
[89.6s] and you can use code for this, you could even use her.
[91.7s] It's for it, although I think if I had a choice out of all these,
[94.3s] I'd use chatchipathy or code at this point,
[97.8s] because it's just much more visual when you build in stuff like this,
[100.6s] you can actually fork the conversation.
[103.4s] So you can click on new over here, and inside the same folder,
[108.4s] you can have agents working in parallel on building out the H&WS.
[112.6s] Now, this is really, really useful.
[114.6s] So for example, when I'm coding out the H&WS,
[117.8s] inside, called what you see on the left hand side
[121.3s] is that I will test and run different implementations side by side at the same time.
[126.4s] What I mean, for example, is that when I was building out the video agent,
[130.4s] I was also working on the H&WS.
[132.2s] And I was having them both being built side by side
[135.0s] inside the H&WS system.
[137.5s] And so the great thing about chatchipathy as well,
[141.3s] is that you can just start a new chat here,
[143.3s] and you can say add inside the H&WS,
[149.0s] an image generator.
[152.6s] And then we'll say use G556 for generating the images
[158.3s] with the OAuth slash CLI setup,
[162.8s] and make sure it actually works, but it's allow me to save anything that I create.
[167.7s] And so what you can see here is that we now have
[171.8s] two conversations at the same time
[174.2s] for our new H&WS that were built in.
[176.6s] So every idea that you have that you think,
[179.6s] wow, this would be valuable, you can build it at the same time.
[182.6s] And this means for example, you can have like six different conversations
[186.4s] inside the same project that would be encoded outside by side.
[189.8s] So we can have, for example, a video agent,
[192.8s] an SEO agent being coded out, the lead generation being coded out,
[196.1s] and we just add in new workflows inside the H&WS system,
[199.6s] as we go along.
[200.9s] So you can see here that it's asking for your questions,
[203.8s] I'm just going to say, yeah, just figure it out and build out, right?
[207.6s] I don't need to give instructions like that to G556,
[213.7s] but this is smart enough to figure out for me,
[215.5s] and if it really is bad then, okay, no problem.
[217.8s] I can go back and fix it later.
[220.2s] The other cool thing about using this is like,
[222.1s] whether you're using, for example, called or using chatGpc,
[225.0s] you can plug in your existing subscriptions into this system.
[229.3s] So if you use Claude Woll, ask it to build,
[232.2s] Claude CLI into the H&WS, and that way you don't need to use API,
[236.4s] which means that you don't blast through loads of tokens,
[238.7s] which means that it doesn't cost you anything extra,
[241.2s] to number one build the H&WS and number two use the H&WS.
[245.7s] Because for example, we have a look inside the system,
[248.9s] we have an agent who goes with Claude with Hermits,
[252.6s] with Vantigravity with Codex.
[254.2s] Now Codex already subscribed to chatGpc before I build this,
[257.2s] and so it didn't cost me anything extra.
[259.0s] Claude already subscribed to it before I build it,
[261.5s] didn't cost me anything extra.
[262.7s] Hermits free open source projects, we plug on the Rue sub-brain into it,
[266.6s] doesn't cost me anything.
[267.8s] So you see how you can build these agente cooperatives systems for free,
[271.1s] that's very important to note as well.
[273.7s] And now this is beginning to code step by step.
[278.4s] It's adding the image generator inside the existing storming home,
[281.4s] this is one.
[282.6s] And by the way, for the actual agent or system that uses the template,
[287.4s] that was all set up and configured here.
[289.3s] So it says the new system has separate configuration
[292.3s] and what I would do if you are installing this template from scratch on the upper floor,
[296.2s] it's just tell it, listen, test everything out, anything that doesn't.
# oracle/transcripts/chunks/c05.mp3
# model: tiny | lang=en
[0.0s] work, tell me and make sure that you authenticate it or if you can't authenticate it,
[5.4s] that you ask me so that I can set out.
[7.8s] So the template set up is pretty much finished at that point.
[10.4s] That's way, way easier.
[11.7s] Whereas for example, be on an agent or a search, you know, you could be here for a few hours
[16.0s] or you know, you might see something that breaks tomorrow and then you have to go and fix it
[19.4s] as well.
[21.6s] Now, when we were talking about the memory system as well, that obsidian system, you can
[26.3s] go back into the agent or else here.
[29.7s] And you could say, also add the memory system from obsidian, so I have obsidian installed
[35.1s] locally and install that into the agent or else as well.
[38.5s] So it's automatically synced and also make sure that the chat activity and hermys are synced
[45.0s] to my memory system.
[46.5s] So when I use them, you update the obsidian as an immatarver.
[51.7s] But also when I ask you questions, you learn from the obsidian vault that we have locally
[57.4s] and you use that as context.
[60.0s] That way everything is personalized inside the H&O S.
[64.0s] So this is the memory system and how you can install that.
[67.6s] And this is pretty cool because now we have the front end plus functionality that's being tested
[75.2s] of it.
[76.2s] We have the custom workflow that we're building out here and we have the memory system that's
[82.8s] being built over here.
[84.4s] So we're adding three different layers as we go along.
[88.2s] We've done that in a space of 27 minutes.
[92.9s] There's going to be some workflows as well over the other ideas for and you're like, there's
[96.2s] no way it could do that.
[97.6s] I would genuinely just ask because most of the time, G556 can handle anything that you've
[103.1s] got out of here.
[104.1s] I've not seen G556 be unable to build this stuff out.
[108.5s] It's pretty unbelievable, everything that you can use here.
[113.1s] So I'm going to say figure this out.
[115.8s] If I had more time if I was here for a few hours or whatever, then what I would actually do is give
[121.4s] it more instructions and be more in-depth and just take a little bit longer to answer these
[126.5s] questions, but I'm trying to save you time and I'm trying to show you in real time how it works
[130.5s] about it being boring, honestly.
[132.6s] So if we have a look over here, we've now set up G556 so we've in-miss generation, we've set up
[138.9s] the front and functionality and we have the dashboard here.
[142.0s] We can also improve it and test out side by side inside chatchubity.
[146.8s] So this is the offering system of built and this is the dashboard that we're built.
[151.8s] So if we go inside the Hermiss agent here, we're just going to say working and make sure Hermiss
[157.5s] agent actually replies to us, boom, and it says yes, I'm working what do you need?
[161.9s] Now this is amazing because now we have Hermiss agent, we have G556 plugged in which is like
[168.0s] one of the best frontier models in the world.
[169.9s] We have chatchubity codex and that's creating some Missist stuff as well.
[175.6s] So if we go to Hermiss agent here, we can use it identically and we can use it inside our H&WS.
[180.8s] Now we've got the inside chatchubity codex and now we have the mission control as well.
[184.7s] If we click on any of these, it takes a straight to the commissation.
[188.0s] So UI looks nice and looks clean and looks simple, which is great as well.
[191.8s] And then over here we have the live activity, so we can see what's being done.
[197.0s] We have the work that moves with recent operations, we have both of the agents that we've built,
[201.4s] and we have everything else ready to go.
[204.6s] Unbelievable how easy it was to set this up.
[207.3s] Now of course we do have the emiss generator that needs to be tested and we'll see how that forms
[211.9s] as well.
[212.9s] And then we've also got the memory system that's being built out as well.
[217.8s] There's a lot of layers do you need to add to this as well.
[220.7s] That might be another question they're asking.
[222.8s] So something that's really, really useful is the conversation history.
[228.4s] Now if I open this up and it just starts from blank every time, not very useful.
[233.0s] But if I can start a new chat, like you can see here, and we can switch between them, that is great.
[238.6s] Because then you're kind of using Hermiss like chatchubity.
[240.8s] So you can go back and forth inside the chat of it, but also you can use it identically because
[246.8s] you've got Hermiss agent, basically have superpowers, you can do computers, you can edit files locally.
[253.0s] And so it's kind of like you've created your own better version of chatchubity, this is chatchubity,
[257.9s] but it can't edit my local files.
[259.5s] Whereas for example if we go over to Hermiss agent, this can work directly here too.
[263.2s] So that's great.
[263.9s] Now one thing that I also add here as well is a system where you can essentially save everything
[272.8s] that you've created.
[273.8s] Now come on to that in a second, but let's have a look at this image generator.
[277.8s] And you see how it would basically build an outer and soft tool.
[281.5s] Think about how amazing that is.
[283.5s] We're at the stage now where you can build your own tools, you don't need any technical knowledge,
[288.8s] you don't need to be really technical or understand HTML or be a coder to build this stuff out.
[295.1s] You can just build your own custom workflows using our automation in a way that is
# oracle/transcripts/chunks/c06.mp3
# model: tiny | lang=en
[0.0s] custom-wise and personalized here. That is unbelievable. So if we go inside here and we're like,
[4.3s] okay, create a image of a cat eating biscuits and we'll delete this bit here, we'll change that.
[11.8s] We'll get rid of that. You see how it's got this aspect ratio as well? So I didn't even ask you to do that,
[17.0s] but code X4 ahead because that asks it to figure out. That's why I quite often prefer just
[21.4s] giving the AI to figure out because it has better ideas and it has more time to give attention to this
[26.7s] than I do. Like if I've been coding this out for hours, I'm going to get tired, I'm going to get
[32.2s] distracted, is going to be difficult to focus at times. You might get frustrated sometimes.
[37.1s] Well, where's if you give the decisions to a frontier model like GPT-5 on 6, it can do it all for
[43.2s] you and handle it as you go along. So now we can say, okay, I want to just go with a cinematic image of a
[49.6s] cat eating biscuits. We've got the prompt, we've got the aspect ratio with cinematic, we'll hit generate
[54.7s] image and that may not work, but so far it's actually looking really good, really impressive actually.
[61.5s] And so one thing that might happen here is if we click from image to video to homeization, I wonder if
[68.0s] it's going to stop? Yeah, see how we click on that and then it's stopped generating. So we need it to be
[73.0s] able to run in the background. That's another great feature. And you'll see the more you test it, the more you'll
[77.4s] see ideas from improve it and more things that will come up that are kind of like bugs by this. And so
[83.1s] what we want to do here is just say, okay, inside the custom workflow image generator and this is exactly
[89.2s] why you want to rename the conversation. So if you right click and rename the conversation, you can see exactly
[93.9s] what you're working on. It keeps makes it easy to keep track. We can go inside here and say, based on the image
[100.2s] you, if I generate an image, can you just make sure it still runs in the background whilst I'm moving around?
[107.1s] And also what we've seen here is that it's actually tested out and it's found some bugs that we didn't
[112.6s] even know about. So it's turned a 500 error and it's uncovered a real bug there which is fantastic
[119.4s] as well because it's still in the bug checking for me. But overall it's looking very, very promising
[125.0s] that's beginning to work. I mean, a bit of we go back to the app here and we're like, okay,
[131.8s] create an image of a cat and then we just hit generate image. That should actually work without us
[137.8s] switching between them. But let's see. So the mission control, we have the different agents built in.
[143.0s] We have the studio of custom workflows as well. But this is beginning to look and form something
[149.8s] that is actually amazing. And you can see here that we've got the agentic memory of all as well.
[157.0s] So this is the system that we've used and as you want to add that as a tab. So I'm going to say,
[163.0s] add the memory system as a tab. So the I can easily see the memory has been added and see what's
[170.0s] been done and also visualized the inside like a beautiful knowledge graph and we can hit into here.
[174.6s] Now it's beginning to run out of work usage as you can see because I've been running on fast mode and we've
[179.8s] created so much stuff started by side. I'm going to say figure this out. But the 8020 of building the
[185.5s] agentic operating system is being done as you can see. And so all of this is working in the background.
[190.5s] You do need to be careful with the token usage as well because you can easily run out of tokens.
[194.6s] If you're wondering how to check your token usage, you can go inside the advanced section here and you can
[200.6s] switch the effort levels and speed and everything else. As you can see how many tokens you've used side by side.
[207.4s] And you can also click on reset usage as well as an option. Now this is beginning to code out each part as well. Nice.
[217.0s] Memory system should be added in and we hit the limit. So there we go. We hit the usage limit already.
[226.2s] We can actually click on reset usage. I'll be quite interested. I know if we can and I'll. Oh yeah we can.
[231.2s] Of course you can do a weekly reset usage limit as well. I didn't know that.
[237.0s] That's pretty good. And then we'll go out weekly usage limit that resets on the 18th which is quite a few days ago.
[242.6s] But yeah and also when you're being out with this. If you want to reduce the amount of tokens
[248.6s] because this can be come up problem especially if you're creating some of the things in parallel using the
[254.4s] GenteColoring system, then you've got a few options. So you can switch your fast mode. You can switch off the
[259.9s] intelligence so you can actually dial down the intelligence level. You can also change and click on advanced
[266.8s] and then you can switch between the models. So if it's just something really basic then you could switch
[272.0s] the model to like 25.5 and that would use less tokens. You can change the effort level to light or medium
[279.0s] and you can change the speed standard. Now when you're building all this out,
[283.4s] the more you reduce the amount of tokens you use by dialing down the effort dialing down the speed,
[288.7s] dialing down the model, the worse the outputs will be. As always that balance with token optimization
[294.6s] creating something that you really like also using the plan.
# oracle/transcripts/chunks/c07.mp3
# model: tiny | lang=en
[0.0s] about running out of credits.
[1.7s] Also, something else that's taken to account here
[4.2s] is that there are loads of different ways
[5.9s] you can reduce the amount of tokens and minimize them
[8.5s] even more.
[9.4s] We've actually got 10 different strategies
[11.1s] inside our token minimization playbook
[13.2s] in the app off of volume.
[14.3s] They can help you save like Nibes,
[15.7s] 95% of your tokens.
[16.9s] So if you want to get almost two to three X,
[19.6s] out of your system and be able to code for more
[23.7s] and use more tokens, then you can use this system here
[26.4s] and it will actually reduce the amount of tokens you use.
[29.2s] And that's also one of the reasons
[30.6s] that we recommend a template to set up
[32.6s] because for example, with the template to set up
[34.5s] as you can see right here, that just installed it,
[36.6s] configure it in the background,
[38.0s] and that was already done in like 10 to 20 minutes.
[40.4s] That is way better than having to build this from scratch yourself.
[44.4s] So it's your choice, but for me personally,
[47.8s] I would prefer to use like a template and set up
[49.9s] just so I can save time because, you know, times is really valuable.
[54.0s] So we have talked about how to build your own agent operating system,
[57.6s] how to add custom workflows,
[59.6s] how to set up the memory inside there,
[62.4s] how to test different works,
[63.9s] how to add new features, and you can see, for example,
[65.9s] we have the image of the cat E in this kit
[68.0s] now as well, so that custom workflow works really well.
[71.3s] One thing that I'd recommend to you all at your time,
[73.7s] so you look at where you spend your time
[75.7s] and then how can the agent OS help you?
[78.4s] So anything that you have an idea for,
[80.5s] you can build it as you see.
[81.9s] So for example, the image studio, we have that idea,
[84.2s] and we build it out.
[85.8s] Now at this point, you might be wondering, okay,
[87.9s] what should you be folks, you know,
[89.4s] what custom workflow should you add to your agent OS?
[92.1s] Well, I would look at where you spend your time
[94.3s] and then work backwards from there.
[96.0s] So, anything that you need to do every single day,
[100.2s] on a schedule that is repeatable,
[103.7s] and doesn't require you to be there, which is basically nine or seven of the stuff,
[108.0s] you should be automating, and you should be building custom workflows.
[111.2s] So let me show you some examples of what we build here.
[113.5s] So inside my agent OS, we have the content pipeline.
[116.5s] The reason that we did that is no one else was in loads of time
[119.3s] on keyword research, and I was using tools that I didn't really like.
[122.6s] So for example, I was using a trusted keyword research.
[125.0s] It wouldn't really give me new data,
[127.4s] and it would give me old suggestions that weren't helpful.
[130.3s] With this system, I can type for example,
[132.5s] agent OS or whatever I'm trying to focus on when it comes to SEO.
[135.4s] It will pull in that data, and it's custom made
[140.2s] to all of my websites and niche them in.
[143.5s] So if we keep scrolling down here, you can see all of the different keywords that we've got.
[148.5s] And so I'd be safe for me about 30 minutes a week.
[151.3s] Then from there, we can click on use topic,
[153.9s] and we can actually create the content and deploy it to our website and one single click.
[158.9s] That is super useful for me.
[160.4s] Now if you're watching this, you might be like, well, I don't do any SEO.
[163.3s] Like, why don't I want that?
[164.7s] And that's exactly why.
[166.0s] You want to go back into GV5.6,
[168.4s] based on what you work on day today, and tell it exactly what you do.
[172.7s] Based on your time audit and ask it to build aint.
[176.4s] So if you spend a lot of time doing research, build aint.
[179.1s] If you spend a lot of time doing lead generation,
[180.8s] add lead generation to all inside that.
[182.8s] That's how it works.
[184.0s] And this just loops around around.
[186.2s] So like, for example, when you start automating all this stuff,
[190.2s] I don't think that you're going to find you at less.
[192.0s] I think that you're going to find you work more productively,
[195.3s] and you achieve way more.
[196.4s] Like, I mean, for me, we can create way more content.
[200.8s] We can recreate more people.
[202.3s] We can grow way faster, because we have all these systems in place.
[206.0s] And so like, for me, I don't see it as like a way to,
[210.4s] three out of my time.
[211.4s] I see it's a way to save time so that I can use that to focus on other parts of the business.
[215.7s] And then just keep cycling through.
[217.6s] Okay, what do you work on?
[218.5s] What do you work on?
[219.2s] What do we need to do next and next and next?
[221.6s] Also, there's some other good really cool stuff that you can build inside here too.
[226.2s] So one thing that we actually did is, we notice,
[229.0s] okay, we have so many different agencies.
[230.8s] We've got a claw to get hermons, we're anti-gravity, we have to do a codex.
[233.8s] Why don't we add those inside a system that we can orchestrate?
[238.4s] And so we actually had a paper clip inside here as well,
[240.9s] so that we can have an organization, a team of AI agents working together.
[245.0s] They can go off and build stuff 24-7.
[247.4s] We can see everything that we've created that is really useful.
[251.2s] And also a lot of people are asking us to build this in.
[253.5s] So for example, in their proper form of community,
[255.8s] when people are asked stuff, we actually build it for them.
[258.0s] Like for example, there was someone was asking this morning about how to use chatchewity in the new app.
[264.5s] So what we did is we recorded like a one-hour course for them,
[268.2s] showing them exactly how to use chatchewity.
[270.8s] And so what I really like to do is get ideas from the community,
[275.8s] figure out the problems they have and then solve those problems with the agent OS.
[280.0s] And I think that's really good for you.
[281.5s] Like you can speed to your team, you can speed to your audience.
[284.4s] You can look at the problems that you have and solve each one of them,
[288.4s] one by one, by building the right systems in.
[291.8s] The same with the group chat here.
[293.0s] So we have a group chat and these are just ideas that were close, you can build in.
[296.6s] Because we have finished building out the experience.
# oracle/transcripts/chunks/c08.mp3
# model: tiny | lang=en
[0.0s] sample project. So this example we actually create a group chat where you can message a room.
[6.4s] And then we can set a keyboard, Hermit, come up with a new game idea to build. It will read
[11.4s] our obsidian vaults of the memory that it tells that earlier. And then all our agents just
[15.5s] chip in and the speed to each other and the bounce ideas off each other. So like for example,
[19.8s] Joe and I are like, oh that's a clever twist. Clawd, let's do that. And then code x as I
[24.5s] know what this sounds really good to integrate over and over and over. Let's add that to our pipeline,
[29.8s] which brings me to the next setup. So we've got a paperclip for agent orchestration. We have
[34.8s] the AI agent group chat. We can also see the history of all our group chats here. Always good to add
[39.5s] history inside the agent or I suppose you won't be able to find stuff in your loop stuff and you're
[43.5s] great. Let's all substitute the confine later. And then finally we have the pipeline. So with the pipeline system,
[49.3s] you can drop an idea in here and you can go from get me idea to going into the human approval
[56.5s] section and just implement it. And this was really good because I found like a lot of people
[60.6s] including myself would have so many amazing ideas but it's very difficult to implement it for him.
[65.4s] So with this system, you can take all your ideas all to master this from the group chat because
[69.8s] the agents add this into the pipeline. And then anything that you like right, we need to build this today.
[74.8s] Okay great, let's just build it. It will come up with a plan. It will plan it out. Then we can
[79.8s] approve it. Then it will build it and then we can view what was actually built previously. And we actually
[85.0s] have a full gallery where we can see all of our ideas. And so we can go from idea to implementations so quickly.
[92.2s] Which is really really cool. So for like okay we need to build out this website, boom shout,
[96.0s] like we can open up that website here later. And again I think it's so important to have like a gallery
[101.4s] and a workspace and stuff that you can see and say for later. Because if you don't,
[106.2s] the biggest issue is that you will not be able to find it. It would just get lots of new downloads.
[111.8s] And that will go with the other 100 projects you're working on and then you won't be able to find it.
[115.7s] So that's why I'm going to a gallery and a pipeline is so so useful. Now what we also have over here is we have,
[122.3s] for example, Claude. And anytime I'm like Claude actually brings something useful out,
[126.9s] we add that feature in as well as a separate tab. I'd recommend that for you too. So for example,
[132.4s] if you saw the ultra-code update, this was amazing. You can have a 14 of agents working in parallel,
[139.3s] working together towards one task and you don't have to orchestrate it, Claude is your extraator
[143.9s] and it builds the subagents for you. So with this system we just added that as a separate tab. And it's like
[149.4s] you create your own custom version of Claude that you can edit, you can change your UI of any time you want,
[154.2s] you can come back to at any point. Then we have Hermes. So Hermes again, really, really powerful.
[162.0s] One of the things that I noticed is like they bring out new, there's always new models that we can plug into
[166.2s] Hermes and this way you want to focus on the system, not the model. So for example,
[171.4s] G556 comes out, no problem. We're going to build that out. It doesn't change how I work.
[176.5s] Doesn't make me have to rebuild anything. It's just like we can add the model inside the system.
[181.4s] The same, for example, a lot of people asking about how to have a voice activated AI agent that we
[185.8s] can talk to. And so we actually build that into this system as you can see. So what it can do is you can listen to me,
[192.2s] you can talk to me, it can build stuff out for me. And then again, we've got the full gallery here of
[197.9s] everything we've created. But we have Hermes all recall. For me, I was just find way too much time was being taken,
[205.0s] research in the latest news. And so I created this. I also think a great tip for you here if you're
[209.5s] building out your own agent or S is that you make sure it's super visual and fun to use. So for example,
[216.0s] we just added this law icon here. I think it's just a little touch. It's a little UI touch, but make it fun and easy to
[223.8s] use and very memorable. So when I see this in my archive, I know exactly what this does because I associate
[230.2s] the image with the workflow here. And if you don't like the UI of some like that, you can always customize
[235.6s] it for later too. So we have an image studio here too, but we actually have a video on a voice session as
[242.0s] well. You'll not see your previous sessions or what's been done across all of your AI agents. Which is
[246.5s] great when you see the status, we can see the skills installed into a Hermes agent, the campband, the plugins,
[252.4s] the doctor of the insights. These are all just for real cool ideas. Again, as you make it more complex,
[258.4s] more things can break. And also if you're using APIs in any part of that, just be careful because
[264.4s] you know, no more than you don't want to rinse the tokens if you're using the API XT in get expensive.
[268.5s] And number two, sometimes the API is for break or let expire and you have to be careful about too.
[274.2s] What I also like to do is let separate the outputs by the system depending on the model. And one of
[280.9s] me and by that for example is we have Hermes here. What I recommend to you is that you build out workspace
[285.9s] where you can track all your creations by the model that's created it. So when we were testing out
[291.8s] Grok 4.5 earlier this week, we have all of the creations from Grok 4.5 saved inside the workspace.
[299.4s] Really is.
# oracle/transcripts/chunks/c09.mp3
# model: tiny | lang=en
[0.1s] We have the MCP section here, we have the actual Hermes dashboard embedded inside the system.
[5.7s] Another really good tip as well is that if you see anything open source,
[10.5s] you think, wow that would be great to build in.
[14.3s] For example, if we have a look through this list, we see what we've got here like on the room.
[19.6s] This is a free, this is a free way, a free API that you can use to build and automate anything with.
[27.3s] Now why is that important for you?
[29.3s] Well, let's say for example, using the H&S and you want to reduce your amount of tokens to use.
[33.3s] Well, you could take a project like this, you can grab the GitHub link, you can go back into chatchipity here,
[39.3s] and you can see it okay, based on this, build this into the H&OS system.
[45.3s] So, I can code for free.
[49.3s] And then that's going to take the details from OmnLyRoot, install it into your H&OS,
[55.3s] and all of the sudden, you've not just got chatchipity in Hermes in there,
[59.3s] you've now got a system with a free code.
[61.3s] And if you're wondering how does that look like in reality?
[64.3s] Here's an example.
[65.3s] So, this is a custom tool we made with OmnLyRoot, where we can code out for free,
[70.3s] we have the chart over here, we have the preview over here, we have the workspace on there.
[74.3s] Really really powerful stuff.
[76.3s] We have for example H&Y3 code, and also some of this stuff will expire.
[79.3s] So for example, H&Y3 is only available for free until the 21st.
[83.3s] And so, no problem, as soon as the 21st hits, we can delete this whole section,
[87.3s] or we can hide it, and we can do something else.
[90.3s] The same for example is to kind of figure.
[93.3s] Like, to kind of figure is an API, like it's not that nice to use,
[97.3s] and it's kind of like you have to find a place to store it, and be able to come back to it.
[101.3s] So, what do we do?
[102.3s] We build out our own custom tool, we have to kind of figure,
[105.3s] where we can see everything that's created, we can chat with it,
[108.3s] we can ask it to do stuff, and everything is stored inside one system right here.
[113.3s] And then anytime something is trending or interesting,
[116.3s] I want to play with, for example, I loop, loop engineer was something big,
[119.3s] that was, you know, everyone was talking about about two weeks ago,
[123.3s] we've loaded in the loop engineering system here.
[126.3s] And then the other thing I would say here is,
[128.3s] when you're building out agent of rent systems,
[130.3s] it doesn't just have to be for business stuff.
[133.3s] So, for example, you could have a system like this,
[136.3s] where you can generate music.
[138.3s] It could be, for example, you goals.
[140.3s] If you were really into more time, you know,
[142.3s] you can have something related to that.
[144.3s] It could be like a track of a training, it's simple like that.
[146.3s] Whatever you prefer to do, you can plug it inside the system.
[150.3s] So, you could, we have a music generator here,
[152.3s] if I want to just create some music,
[154.3s] I'm interested in for particularly for like,
[156.3s] ambium music, then I can just plug this inside the system,
[159.3s] we can get files, we can come back to the AI music we've generated later,
[163.3s] and it's already to go inside there.
[165.3s] Inside there is one.
[167.3s] So, those are all the systems that we get.
[169.3s] Another orchestration system, as well,
[171.3s] and this is with Hermes, is the Kambam board.
[174.3s] Now, with the Kambam board, what we have is the system
[177.3s] where we can drop in a task, it gets automatically triage,
[180.3s] and then all of that agents,
[182.3s] and we've got many different agent profiles,
[184.3s] get signed to that task as well.
[186.3s] So, yeah, it's unbelievable,
[188.3s] it's really changed the way that I work,
[192.3s] and the other thing I would say here is it changes all your team works as well.
[195.3s] So, if you have a team,
[197.3s] like, for me I have an SEO agency,
[200.3s] you know, we've got a team who needs to be able to use AI and that sort of thing,
[204.3s] we can give this to like our team,
[206.3s] that it is, and they can use the video agent.
[208.3s] And if we have a look down here as well,
[210.3s] these are all reviews and testimonials from the Air for Boarding,
[213.3s] you can see here, for example, like, so many people,
[215.3s] again, awesome wins and awesome results with the agent or S like Jacob,
[219.3s] he said it up in about 10 minutes and you saw that I said it up today as well.
[222.3s] And Jacob was spending like 12 hours a day trying to build out the agent OS,
[227.3s] then he just did it in 10 minutes by using our template.
[230.3s] David was saying, like,
[232.3s] Julian is cooking hard on his agent OS and anybody else seen.
[235.3s] So, it's a great way to customize it.
[237.3s] And you can see, for example, like,
[239.3s] sheena as well, sheena actually created her own version of our agent OS,
[244.3s] and used it for lead generation.
[247.3s] Now, one thing that I guess asked all the time by people was like,
[252.3s] you know, what, if you're, for example, thinking about selling AI agent services or AI automation,
[258.3s] what's the one thing the clients want to see?
[261.3s] What is in demand right now?
[263.3s] I'm for sure at the Asian Opera Insisting, like,
[266.3s] if you can build out custom workflows that solve all the problems your customer has,
[271.3s] by creating amazing systems like you've seen today,
[274.3s] so if they need help generate images, you create the image studio.
[277.3s] If they need help generate videos, you create the videos you,
[280.3s] if they need help generate social media content, you create a social media agent,
[283.3s] and how'd you do it?
[285.3s] You just go back into codex and you test it out for later.
[288.3s] So, you can easily build out all this stuff and you just grab their idea,
[293.3s] you solve all the problems you go from there.
[296.3s] Now, as well, one thing that I'll say,
# oracle/transcripts/chunks/c10.mp3
# model: tiny | lang=en
[0.0s] He was like, you might be tempted to have Claude and Codex work on the same thing at the same time.
[7.0s] I personally wouldn't recommend that.
[9.5s] So if you are building out with an agent and you're trying to work on the agent,
[13.8s] as you're trying to improve it day by day, trying to make it better as a system,
[18.0s] the thing that I'd recommend here is that you just use one agent.
[22.0s] So for me, for example, I'm using Claude right now and I use Faber 5 to help me,
[26.0s] you might prefer using chatchipy to you might prefer using Hermits, but seek to one.
[30.0s] Because if you have multiple agents, it multiple different apps, number one that gets really messy when you're building out an agent offer in system.
[38.0s] And number two, they could like overwrite or delete something that the other agent is building.
[45.0s] So for example, Claude is building out something that's just broken.
[48.0s] Chatchipy might be like, let's remove that from the code.
[51.0s] And then all of a sudden you've got two agents kind of interrupts any chance work and that can get super messy.
[56.0s] Wouldn't recommend that.
[58.0s] One thing that I haven't tested, that would be very interesting in testing mode, is with Codex is very easy to add plugins.
[66.0s] Now, what would be very interesting here is if you could add those plugins inside the agent or S as well.
[73.0s] So that's an example that we've got email and Gmail plugged into Codex, right, into Chatchipyty work.
[80.0s] So you can send the email address, you can manage your email.
[83.0s] So you can draft replies for us and if it else.
[85.0s] Now, what would be very interesting is say to Codex, a build out this scheduled task into my agent or S, so that we have the email agent inside there.
[99.0s] I'm not going to do that right now, because I think it's a task that we'll probably take about 20 or 30 minutes to really perfect.
[104.0s] But that would be interesting to take for the plugin section from Codex, and start adding that into your agent or S, so that you can automate and build anything as you want right there.
[115.0s] Other things that I would recommend.
[118.0s] Number two is you might be tempted to put this on a website, for example, like a lot of our websites we use with NetLefite.
[127.0s] I wouldn't do that just because again, like I think you need to keep it secure in everything else.
[132.0s] Also, if you want to reduce the amount of tokens, you can also use stuff like Free APIs and you can use local models.
[140.0s] So if we go to the local section here, we can use local models from Olammer or LM Studio, and this is particularly good for like the tasks that a quite time consuming, but don't need frontier levels of effort.
[155.0s] So let's say for example, we're looking at all of the models on Olammer.
[161.0s] I'm like, okay, or NetLef, or NetLef looks awesome, right?
[165.0s] Actually, it's a really good tool. I've test it out.
[167.0s] Persona, we have the local lead board on Goldiebench.
[171.0s] We've all the local models, and you can see what we've created in T.R. web, Formula Finals.
[175.0s] And so you might be tempted to build that into your agent or S as well.
[179.0s] And I think it's actually pretty interesting to build out a local code.
[183.0s] And so what we have over here is we have this local section, and we have the preview of F and created what the works face of our previous generations, and we can use local models in SOT System 2.
[195.0s] And that's another way to use it for free, because then you are a combination of like Omniru, which is free way to use like 90 models, and it kind of orchestrates across them.
[204.0s] You have HY3, which is another free API, you can plug into the system, and then you have local models, which are basically free forever once you've installed them.
[212.0s] So between those two things, those free things, you can really reduce the amount of tokens the years as well.
[219.0s] What else can we do here? The other thing that I like to do is when I'm testing out all of the models, with the same agent, the builds out the HNOS, with all the creations I just put them on Goldiebench so that I can document the process here.
[234.0s] There's another good way to generate leads, because people are searching for interesting stuff, people aren't in no what you've built with it.
[239.0s] People are interested in your agent or SOT if you actually share what you're working on and everything else, especially if you're working on some new unique stuff.
[246.0s] So you can actually give your agent or S access to NetFies well, and then it can document the whole process, the whole building of the agent or S on what you've done.
[256.0s] Kind of like a journal, but it's not you update, and it's not you writing this agent, which is really cool to us.
[262.0s] So that's basically it. That is the whole system. How to build an agent or for an assistant?
[267.0s] How to customize it as much as you want? How to add a memory inside there? Which agent do you use to build it?
[272.0s] How to reduce the amount of tokens you've won. Also, how to create a system that actually works rather than just breaks?
[280.0s] Also, how to use it in a simplified way using our templates. Now if you want to get my full agent or for an assistant, you can get that inside there,
[287.0s] or for boarding inside the Aginal S system section right here. So we update the stately and get the set file for it and you can get the last, you see the last update date here.
[295.0s] You can get a full guide and how to use it, and then every time we build something new.
# oracle/transcripts/chunks/c11.mp3
# model: tiny | lang=en
[0.0s] is full. So for example, like we create a chat GPT 5.6.1 out course, you can get that over here as well.
[6.0s] And then also if you want more help and support on building out with the H&O S, you can post inside the community
[12.0s] and we reply to you and create video tutorial for you ever single day. So for example, like Matt was asking about Ome and the best way to use it and what you can use instead,
[20.0s] we solve that problem for him and we reply to everyone and just help each other as much as you can in the best way that it can as well.
[27.0s] Inside the classroom you get access to all of our best trainings. If you're a complete beginner, you can go from begins to expert inside this section.
[34.0s] And inside the calendar, you can jump off weekly coaching course. Now this is super useful because you can share your screen.
[40.0s] You can ask questions you can get help and support him. And you can also meet other community members who are building out this sort of stuff as well.
[48.0s] Inside the map is well, you can meet people locally who are building out with AI agents like you and the soul available inside the AI Prof.
[56.0s] or link in the comments description or just go to the AI Profible.com.
[60.0s] One other thing that has been really useful when we're building out all this stuff is that when a model, like for example, doesn't work the same way or you can't use it anymore.
[72.0s] So for example, if a great example now, because you've got a system that improves everything with day, it doesn't really matter about that sort of stuff.
[80.0s] You don't need to worry about that. So if GVT5.6 comes out, that's great. We can add in.
[85.0s] But if, for example, Fable5, I've got taken down, which you did a few weeks ago, it doesn't matter because in like two minutes you can go back into chatchipity and you can say,
[96.0s] you can make remove the integration with Fable5, add this in, etc.
[101.0s] And also one thing that's really useful is because you've named the conversations, you can come back in like two weeks and edit that conversation.
[109.0s] So for example, what I like to do is I applaud if I'm scrolling down. If I need to improve the setup that we have with Pony's now, which is another AI project, I can go back inside the chat here.
[120.0s] And I can say, hey, add this, change that. And because everything is named, it's super easy to organize and it's easy to see everyone as well.
[128.0s] So my very interesting thing is that we've set up two in the past is like computers with the agent OS. So I can go inside her, for example.
[136.0s] And I can say, hey, Hermes, can you just open up Google for me and that sort of thing?
[142.0s] You can actually do that with Apollo, which is the voice agent and that actually has access to computers as well. So it can edit local files, it can edit.
[150.0s] You can open stuff inside your browser and everything else as well. Which is really, really good.
[155.0s] And that's pretty much it when it comes to the whole system. I mean, it's so much fun to build. It's so inspiring to build. It's also great to just build something that feels like your own.
[166.0s] It's never been more fun to use than I think this is only been possible this year. Like, if you went back last year, it was great using AI, but you couldn't build in the same way, but models would just not add this level.
[179.0s] Whereas when you're building the agent offerings, it's now any custom workflow that you have, things that you've never even dreamed of creating before, you can just add inside this, you know, you can plug in like, right, when I just want to delete that, no problem at all.
[194.0s] So it's so much fun, so much fun. So more things that you can do as well is you can get skills and then you can plug them into the system.
[203.0s] So for example, like there's a skill for reducing tokens, which is called Caveman, and with Caveman, it basically reduces the amount of output tokens you plug in.
[212.0s] And you could add that inside the system as well. The final thing that would say, and this is a little feature is that you can delete stuff, give yourself the ability to delete stuff.
[221.0s] So you can go into chat, cheapety, and you just say, okay, make sure that everything that we create inside the agent offerings system, we can delete so that I can remove anything that doesn't work.
[230.0s] Because like, sometimes you might generate something with a gelinfoam to, but then it actually doesn't work very well.
[236.0s] And so how do you get round that? Well, you can just make sure that you have the ability to delete that and just add a delete icon for later.
[244.0s] So that is basically it. That is a full one hour course in how to build agent offerings system, how to automate anything, how to build your own custom workflows. If you want my full setup, you can get that inside the air, pop water and you can get the amazing community that's all using this sort of stuff.
[259.0s] You can connect me personally, you can direct messry anytime you need anything. And thanks so much for watching. I'll see you in the next one. Cheers, Baba.
transcripts/ep_build_agent_os.txt
+78
View File
@@ -0,0 +1,78 @@
=== chunk 00 ===
loading tiny on cpu...
transcribing (incremental write)...
...25 segments, 65s
...50 segments, 131s
...75 segments, 199s
...100 segments, 290s
DONE segments=103 -> oracle/transcripts/chunks/c00.txt
=== chunk 01 ===
loading tiny on cpu...
transcribing (incremental write)...
...25 segments, 134s
...50 segments, 295s
DONE segments=50 -> oracle/transcripts/chunks/c01.txt
=== chunk 02 ===
loading tiny on cpu...
transcribing (incremental write)...
...25 segments, 231s
DONE segments=32 -> oracle/transcripts/chunks/c02.txt
=== chunk 03 ===
loading tiny on cpu...
transcribing (incremental write)...
...25 segments, 111s
...50 segments, 210s
DONE segments=73 -> oracle/transcripts/chunks/c03.txt
=== chunk 04 ===
loading tiny on cpu...
transcribing (incremental write)...
...25 segments, 59s
...50 segments, 135s
...75 segments, 220s
...100 segments, 292s
DONE segments=101 -> oracle/transcripts/chunks/c04.txt
=== chunk 05 ===
loading tiny on cpu...
transcribing (incremental write)...
...25 segments, 93s
...50 segments, 197s
...75 segments, 295s
DONE segments=75 -> oracle/transcripts/chunks/c05.txt
=== chunk 06 ===
loading tiny on cpu...
transcribing (incremental write)...
...25 segments, 138s
...50 segments, 289s
DONE segments=51 -> oracle/transcripts/chunks/c06.txt
=== chunk 07 ===
loading tiny on cpu...
transcribing (incremental write)...
...25 segments, 54s
...50 segments, 117s
...75 segments, 186s
...100 segments, 256s
DONE segments=112 -> oracle/transcripts/chunks/c07.txt
=== chunk 08 ===
loading tiny on cpu...
transcribing (incremental write)...
...25 segments, 122s
...50 segments, 264s
DONE segments=56 -> oracle/transcripts/chunks/c08.txt
=== chunk 09 ===
loading tiny on cpu...
transcribing (incremental write)...
...25 segments, 97s
...50 segments, 165s
...75 segments, 232s
DONE segments=94 -> oracle/transcripts/chunks/c09.txt
=== chunk 10 ===
loading tiny on cpu...
transcribing (incremental write)...
...25 segments, 132s
DONE segments=48 -> oracle/transcripts/chunks/c10.txt
=== chunk 11 ===
loading tiny on cpu...
loading tiny on cpu...
transcribing (incremental write)...
...25 segments, 194s
DONE segments=32 -> oracle/transcripts/chunks/c11.txt
transcripts/ep_hermes_cloud.clean.txt
+55
View File
@@ -0,0 +1,55 @@
# NEW Hermes Agent Cloud Update is INSANE! — AI News Today | Julian Goldie Podcast
# Transcript (cleaned) — 8:14 episode
# Source: https://open.spotify.com/episode/5V241PvCyCee5hctvuNLFn
# Resolved RSS: https://feeds.transistor.fm/ai-news-today-julian-goldie-podcast
# MP3: https://media.transistor.fm/e9f58e1f/568e5431.mp3
# ASR: faster-whisper tiny/int8/beam1 (CPU). Proper nouns restored from raw.
[0:00] Hermes Cloud has just dropped, and this is a really easy way to basically set up a Hermes agent in the cloud. Have it running 24-7. It takes a couple clicks, and then 60 seconds later your agent is live. So you can build an AI employee in 60 seconds. In fact, we call this the "60-second AI employee."
[0:19] So these researchers just put Hermes in the cloud. This is an always-on agent that never sleeps. Members have finished learns, and it researchers, and it can reach you on every single platform. I actually set it up as a profile inside the Agent OS. I'm going to show you exactly how to get the most out of this.
[0:36] Now, if we have a look — for example, we've got Hermes agent built inside an Agent OS. And we can chat with it over here. It's very quick and easy to use. And the thing with Hermes right now is, there's a bit of a "closed-lid problem." So imagine hiring the perfect employee — they work brilliantly right up until you close your laptop. Lid down, it stops mid-task, mid-sentence, everything halts. That's basically every agent.
[1:00] And the problem is that it can get stuck, kicked, and break. If it's not in the cloud, then you can't run AI APIs. And also, if you're not behind your computer and you need to fix something, that can be a bit of a problem.
[1:14] Now what you can actually use is this cloud setup. And you might say cloud sounds expensive and technical. Actually, you just pick a model — it's online in seconds. And you can also set it up with free models as well. So there's a doorway to an agent that never clicks, clocks off. And it can basically speak to you on every single channel that you actually use.
[1:38] So the doorway is the new portal. It's one login that serves over 200 models plus a tool gateway — web search, image generation, etc. Then you've got the agent itself, which is a cloud agent. So you can give it a model, pick a model, and it gets spun up in a hardened container, which is also good. Because then it's sandboxed and protected in the cloud. It's got its own persistent workspace, and it runs 24-7 on the new portal servers instead of your Mac. So when it's idle, the meter only ticks when it's actually working. So if you stop it, then it's not going to use anything.
[2:08] And it reaches across every channel. So you have one cloud agent, one memory, reachable from Telegram, from Discord, from your Agent OS and us. And that's how the whole system works.
[2:20] Now you might wonder, what happens when you actually deploy one? So it's pretty simple and easy. Like, you just name it, you pick a model. Then it gets put inside a sandbox, the dashboard goes live. It works whilst you sleep. And for example, we can give it a task over here, and then it can work in the cloud completely in a separate container. So it's always on.
[2:41] So if you're wondering how to deploy it, you can just go to the cloud over here, and then you just sign in and set it off. And at least it takes like two minutes to set up. It's pretty simple and easy. So there is a cloud portal here, and then you can set it off from there. It's like two clicks. I've already set my agent OS up — I'll currently show you. But yeah.
[3:03] So the cool thing about this is like, every time I was running Hermes, you know, sometimes it would stop. If it was running on local models, you could close your computer and it'd stop running. Also, for example, sometimes something would break inside the Hermes setup, or the API would run out. And it's like, okay, well, how do I fix it? I have to go back to my computer. I have to set it up again. Whereas with this setup, we've got our agent running in the cloud, and it's ready to go. It's pretty fast as well. That's a cool thing. And you can get this set up in like 60 seconds, right? So 60 seconds, you set up your AI employee — then it never asks for a day off.
[3:39] Now, if you're wondering how to build the setup here, we can use a Hermes cloud profile. And then we can plug it into your Agent OS. So this is what it looks like when you're setting it up — you can do it yourself inside the cloud portal. And then you can create an isolated profile as well.
[3:57] Now, you might say at this point, okay, I already have Hermes on my Mac. Why bother with the cloud? Because your Mac can break. Sometimes you want a Hermes agent that's sandboxed in a separate container, isolated, so that's safer. And then also, if you close your Agent OS, you can still run in the background as well. So it's pretty powerful. You might also say it doesn't use a lot of tokens — but you can set this up with a free API, or you can use a free model or a cheap model. And that way, it's better.
[4:26] Now bear in mind, like you can have Hermes running in the background. So even if you don't have a cloud setup, it can run in the background. But the thing is, then you have to go back into your actual device to fix it and edit it. And if it stops working, then if you're out and about, you can't go and fix that. Whereas with the cloud model, you can just log into the portal online. It's in the cloud and you can edit it. And also, some people don't want to install Hermes agent on their device, because they've got files they don't want the agent to touch. And so if that's the case, then instead of what you can do, it's just have it in the cloud.
[4:58] So let's test this out now. So this is Hermes Cloud. We're going to say, okay, come up with some cute [good] ideas for your automation. It's going to start building here. The good thing as well, when you're using Hermes — if you use it inside an Agent OS like this, then it's plugged into your memory. So it understands you, your business, your memory, and everything else, everything inside one place, so that it can use after-context and come up with better ideas.
[5:23] Also, one thing I'll say is it's very, very fast to respond. Way faster than a lot of APIs. For example, when I'm using like Roo Code 4.5 — Roo Code 4.5 took like a minute to do something like this. Whereas with the cloud setup, it seems very, really fast and responds quickly, which is awesome as well.
[5:46] So that's basically it. That's the whole set up. That's how to use Hermes Cloud, how to get it set up. Again, if you're wondering how to get it — you just go to [hermes cloud URL]/cloud. Why would you use it? Because it's always on. You can run with over 200 models. It's easy to set up in like 60 seconds. And also, if you're not technical, then you might not want to set up Hermes itself. And so if you just use the cloud instead, it's easier. And also it can work overnight, even whilst your machine is off and everything else. Plus, there's no service that you need to mess around with. And it's a really easy way to just get the most out of [it].
[6:25] So that's basically the whole system. What did you learn today? You've got an agent that never sleeps, you only pay when it works, you can reach it everywhere, you can try it for free first as well. You can learn how to put it inside your Agent OS. It has a memory that compounds and improves. And you now have a 60-second AI employee that you can use with Hermes.
[6:45] So the cool thing about this — well, we've actually built it into our agent operating system. And we've got a whole section for Hermes. So for example, we have Goal mode over here, we have chat-to-talk, we have a Hermes voice agent that we can control with our voice and get it to build things. And we can see over the history everything that we've built over here. They've got Hermes Oracle, it pulls in the latest news on a daily basis and gives us the most trending news. We have Hermes [Astro] that looks at our competitors and the keyword research. We have an outreach tool with Hermes agent that can do outreach and find leads and generate leads and send emails for us. Basically every powerful feature that's inside Hermes, plus some custom workflows we've built into this system.
[7:34] So if you want to get that — since I've got the AI Profit Boardroom — plus we have loads of training and courses on Hermes agent, you can get that in there. I've got a profit link in the comments description, or go to the app on [profit.com]. And inside that community, you can ask questions, get help — so you can post questions inside the community here. And I personally answer them with daily video tutorials. Inside the calendar, you can jump on weekly coaching calls, get help and support. Inside the classroom, you can actually get access to all of our best trainings. If you're a complete beginner, you can go from beginner to expert over here.
[8:01] If you want to get our new daily updates, you can get over here. If you want to get our Agent OS with Hermes and Hermes Cloud built in, you can get that here with a video tutorial, the last update date, a full guide, and a zip file to install. And you can also meet people in your local area who are using AI agents like Hermes and others. So feel free to get that, link in the comments description, or go to the app on [profit.com].
---
NOTE: Light ASR (tiny) left a few ambiguous terms. Best-guess restorations (flagged):
- "Meas Cloud" → Hermes Cloud; "HGNOS" → Hermes; "anopsite" → agent OS; "SNF" → OS; "zid file" → zip file.
- "Roo Code 4.5" — the slow coding agent he contrasts with cloud speed (uncertain; could be another tool).
- "[hermes cloud URL]/cloud" — the actual domain was garbled in audio ("search.com"); likely hermesagent.com/cloud. Unverified.
- "Hermes [Astro]" / "Hermes Apollo voice agent" — product names unclear in audio; left as best-effort.
- "profit.com" / "AI Profit Boardroom" — his community; the exact app domain was garbled ("forborn.com").
Raw file (verbatim ASR): ep_hermes_cloud.txt (same dir).
transcripts/ep_hermes_cloud.mp3
BIN
View File
Binary file not shown.
transcripts/ep_hermes_cloud.txt
+105
View File
@@ -0,0 +1,105 @@
# oracle/transcripts/ep_hermes_cloud.mp3
# model: tiny | lang=en
[0.0s] Meas Cloud has just dropped and this is a really easy way to basically set up Hermes agent in the cloud.
[7.3s] Have it running 24-7. It takes a couple clicks and then 60 seconds later your agent is live.
[13.6s] So you can build an AI employee in 60 seconds.
[16.3s] In fact, we call this the 60 second AI employee.
[19.6s] So these research just put Hermes in the cloud.
[22.1s] This is an always on agent that never sleeps.
[24.7s] Members have finished learns and it researchers and it can reach you on every single platform.
[29.4s] I actually set it up as a profile and why the Intuit agent OS.
[33.4s] I'm going to show you exactly how to get the most out of this.
[36.1s] Now, if we have a look for example, we've got Hermes agent built inside an agent OS.
[41.7s] And we can chat with it over here. It's very quick and easy to use.
[44.7s] And the thing we've Hermes right now is like, there's a bit of a closed lid problem.
[50.4s] So imagine hiring the perfect employee. They weren't brilliantly right up until you close your laptop.
[55.4s] lid down the stop mid-task mid-sentence everything will type.
[58.3s] That's basically every agent.
[60.5s] And the problem is that you can get start, kick and break.
[65.4s] If it's not in the cloud, then you can sometimes run AI APIs.
[69.9s] And also, if you're not behind your computer and you need to fix something, that can be a bit of a problem.
[74.2s] Now what you can actually use is this cloud setup.
[76.6s] And you might say cloud sounds expensive and technical.
[81.0s] Actually, you just pick an aim in the model. It's online in seconds.
[84.0s] And you can also set it up with free models as well.
[87.3s] So there's a doorway with an agent that never clicks clocks off.
[93.4s] And it can basically speak to you on every single channel that you actually use.
[98.0s] So the doorway is new portal. It's one login that serves over 200 models plus a tool gateway,
[103.6s] web search, image generation, etc. Then you've got the agent itself, which is a cloud agent.
[108.7s] So you can give it an aim pick a model and it gets so in hard and container, which is also good.
[113.2s] Because then it's sandbox and protected in the cloud.
[115.8s] It's own persistent workspace and it runs 27 on new portal servers instead of your Mac.
[121.0s] So when it's idle, the meta only ticks when it's actually working.
[126.2s] So if you stop it, then it's not going to use anything.
[128.5s] And it reaches across every channel. So you have one cloud agent, one memory,
[133.7s] reachable from telegram, from discord, from your agent or SNF and us.
[138.4s] And that's how the whole system works.
[140.8s] Now you might wonder, what happens when you actually deploy one?
[147.1s] So it's pretty simple and easy. Like you just name it, you pick a model.
[150.3s] Then it gets put inside a sandbox, the dashboard goes live. It works whilst you sleep.
[156.2s] And for example, we can give it a task over here. And then it can work in the cloud,
[161.0s] completely in a separate container. So there's always on. So if you're wondering how to deploy it,
[165.8s] you can just go to the cloud over here and then you just sign in and set it off. And at least you
[169.8s] take like two minutes to set up. It's pretty simple and easy.
[174.3s] So there is a cloud portal here and then you can set it off from there. It's like two clicks.
[179.8s] I've already set my anopsite. I'll currently show you. But yeah.
[183.0s] So the cool thing about this is like every time I was running the HGNOS,
[186.7s] you know, sometimes it would stop. If it was running on local models, you could stop your computer and
[192.6s] stop running. Also, for example, sometimes something would break inside the Hermes set up,
[197.8s] or the API would run out. And it's like, okay, well, how do I fix it? I have to go about to my computer.
[202.0s] I have to set it up again. Whereas with this setup, we've got our agent running in the cloud,
[206.6s] and it's ready to go. It's pretty fast as well. That's a cool thing. And you can get this setup in like
[211.8s] 60 seconds, right? So 60 seconds, you set up your AI employee. Then never ask for a day off.
[219.0s] Now, if you're wondering how to build the setup here, we can use a Hermes cloud profile.
[226.3s] And then we can plug it into your agent OS. So this is what it looks like when you're saying
[232.0s] it, you can do it yourself inside the cloud portal. And then you can create an isolated profile as well.
[237.2s] Now, you might say at this point, okay, I already have Hermes and my Mac. Why bother with the cloud?
[241.4s] Because your Mac can break. Sometimes you want a Hermes agent that's sandboxed in a separate container
[249.7s] isolated, so that's safer. And then also, if you close your agent OS, you can still run in the
[255.9s] background as well. So it's pretty powerful. You might also say it doesn't use like a lot of tokens,
[261.5s] but as you can set this up with a free API, or you can use a free model or a cheap model. And that way,
[266.8s] it's better. Now bear in mind, like you can have Hermes running in the background. So even if you don't
[272.6s] have a cloud setup, it can run in the background. But the thing is, then you have to go back into your
[277.5s] actual device to fix it and edit it. And if it stops working, then if you're out and about,
[283.4s] you can't go and fix that. Whereas with the cloud model, you can just log into the portal online.
[288.4s] It's in the cloud and you can edit it. And also, some people don't want to install Hermes agent on their
[293.4s] device, because they've got files, they don't want the agent to touch. And so if that's the case,
[298.5s] then instead of what you can do, it's just having in the cloud. So let's test this out now. So this is Hermes
[303.4s] cloud. We're going to say, okay, come with some cute ideas for your automation.
[309.4s] It's going to start building here. They were good thing as well. When you're using Hermes,
[313.1s] it's like, if you use it inside an agent OS like this, then it's plugged into your memory. So,
[317.7s] understands you, your business, your memory and everything else, everything inside one place, so that it can
[323.9s] use after context and come up with better ideas. Also, one of what we'll say is it's very, very fast to respond.
[329.4s] Way faster than a lot of APIs. For example, when I'm using like, rock 4.5,
[334.5s] rock 4.5 was took like a minute to do something like this. Whereas with the cloud setup, it seems
[340.2s] very, really fast and response quickly, which is awesome as well. So that's basically it. That's a whole
[346.1s] set up. That's how to use Hermes cloud, how to get it set up. Again, if you're wondering how to get it set
[352.7s] you just go to search.com for slash cloud. Why would you use it? Because it's always on. You can
[359.7s] run with over 200 models. It's easy to set up in like 60 seconds. And also, if you're not, if you're not
[365.5s] technical, then you might not want to set up Hermes itself. And so if you just use the cloud instead, it's easier.
[370.6s] And also it can work on like, job overnight, even whilst your machine is off and everything else. Plus,
[375.0s] there's no service that you need to mess around with. And it's a really easy way to just get the most out of
[381.2s] so that's basically the whole system. What did you learn today? You've got an agent that
[385.4s] never sleeps, you only pay when it works, you can reach it everywhere, you can try it for free first as well.
[393.2s] You can learn how to put it inside your agent OS, has a memory that compounds and improves. And
[400.3s] you now have a 60 second AI employee that you can use with Hermes. So the cool thing about this
[405.2s] well, we've actually built it into our agent operating system. And we've got a whole section for Hermes.
[410.8s] So for example, we have Goal mode over here, we have chat to talk, we have a Hermes Apollo voice agent
[416.4s] that we can control of our voice and get it to do so for build things. And we can see over a history and
[420.5s] everything that we've built over here. They want to have Hermes Oracle, it pulls in the latest news on a daily
[424.8s] basis and gives us the most trending news. We have Hermes Astros that looks our competitors and
[429.5s] the keyword research we have an outreach tool with Hermes agent that can do outreach and
[434.2s] fine leads and generate leads and send emails for us. Basically every powerful feature that's inside Hermes,
[439.7s] plus some custom workflows will build into this system. So if you want to get that, since I've the AI
[443.9s] Proford Boarder in plus we have loads of training and courses on Hermes agent, seeing get that in there,
[448.7s] I've got a problem, link in the comments description or go to the app on forborn.com. And inside
[452.7s] that community, you can ask questions, get out of the small, so you can post questions inside the community here.
[458.2s] And I personally answer them with daily video tutorials. Inside the calendar, you can jump
[462.9s] four weekly coaching calls, get help and support inside the classroom. You can actually get access to
[467.8s] all of our best trainings. If you complete beginner, you can go from beginner to expert over here.
[472.6s] If you want to get our new daily updates, you can get over here. If you want to get our agent or
[475.9s] our system with Hermes and Hermes Cloud built in, you can get that here with a video tutorial, the last
[481.1s] update date, a full guide, and a zid file to install. And you can also meet people in your local area
[486.7s] who are using AI agents like Hermes and others. So feel free to get that, link in the comments description
[491.7s] or go to the app on forborn.com.
transcripts/ep_obsidian_omi.clean.txt
+53
View File
@@ -0,0 +1,53 @@
# Agent OS + Obsidian + Omi Is INSANE! — AI News Today | Julian Goldie Podcast
# Transcript (cleaned) — 8:27 episode
# Source: https://open.spotify.com/episode/47Emp3Rnz0DQLj0FfRlHTl
# Resolved RSS: https://feeds.transistor.fm/ai-news-today-julian-goldie-podcast
# MP3: https://media.transistor.fm/51d728ef/8613bf18.mp3
# ASR: faster-whisper tiny/int8/beam1 (CPU). Proper nouns restored from raw.
[0:00] This is a powerful way to have a memory system, as you can see right here. That just works, and it has a powerful set of context, and basically this trains all of our agents on exactly what we've been working on. So that when we plug them all together, for example, all of our own agents — ChatGPT, Claude — understand Hermes. All of our agents work together, and they never forget context again.
[0:28] Now, at the same time, we also have NotebookLM. NotebookLM is one of the most powerful tools that I've ever seen for creating research. And also the amazing thing about NotebookLM is, like, you can generate, for example, videos in a single click for free using this system as well. So for example here, you can see all of these videos that we have built, and these are free videos. We can generate fully educational, fully set-up, fully automated videos in one single click, and we can use NotebookLM to do that.
[1:01] Now when you combine the two — both of these free tools, NotebookLM, and then Obsidian as well — you get a powerful second brain that can basically build and automate anything together. So you might be wondering, okay, why would you want to use something like Obsidian? Well, for example, if I go into Claude here and I'm like, "what did I work on yesterday?" — it's not going to have much of an idea of what I'm doing, what I'm working on, what's useful, what's not useful, etc. And so what you want to do instead is have all of your agents in one shared memory system, as you can see — where you've got Hermes, you've got Claude, you've got Antigravity, you've got, for example, Codex — and they all operate from the same shared memory. That means all of your agents understand each other.
[1:38] You might say, why would you do that? Well, I call this the Infinite Context Engine. It's one memory that every agent shares, and your chat trains your vault, your vault trains your agent, and this is a loop that gets smarter forever. So all of your memories link to, for example, all of your goals, your projects, your areas, your daily notes — and it just links beautifully.
[1:55] The problem with this, if you use something like ChatGPT, is: imagine if you hired an assistant every morning, they wake up, they remember nothing about you, you explain your whole life again — your name, your business, your goals, your clients — every day. Well, that's essentially what your agents are like right now, because they don't have a memory system; your AI is genius, but it has amnesia. Every chat you start from zero. The Infinite Context Engine that I'm showing you today breaks this cycle for good.
[2:23] You might say, well, AI memory sounds complicated and technical. All this is, is a bunch of Markdown files organized into a nice system. That is literally it. This is just a bunch of files and folders that all link together naturally to train your agents on all of your context. And you don't even need coding to do this. You can get your agents to automatically update it for you, and you can get your agents to automatically come up with the ideas for it.
[2:55] Now, there's three tools with one shared brain that I use to set this up. So we have Omi, we have Obsidian, and we have the AI. Omi is basically a system where we can automatically record our screen, our microphone — it's looking at my device all the time, taking an example and looking through all of what we're doing day to day. And then you can see here, it takes notes on what I'm doing, what I'm working on, and it does this like every hour.
[3:17] When we're doing this, this actually exports into Obsidian. Obsidian is where we have our knowledge graph — everything linked together, all of our Markdown files. So what you end up with is Omi creating the memories and taking the memories automatically for you, and then that plugs into Obsidian. And Obsidian is where all your agents come together. If we click on this, you can see it has all of the information about this particular topic, neatly organized, formatted nicely, everything links together. It's just like a Wikipedia, but for agents — everything links together, and you've got everything in one place.
[3:55] So we've got Omi taking the notes. You've got Obsidian gathering the notes that you've created and organizing them. And then that plugs into an agent operating system — which is where you have Claude, you have Omi, every CLI all pointed at the same vault. So every one of them wakes up already knowing you, with one memory in every agent.
[4:16] You might say, well, this works with one tool but not all of mine. But the vault is just text files. Any AI that's trained to read a file path can read your memory. So it's one setup — Omi, every tool with the same brain working together. That's how it works: you capture with Omi, you organize with Obsidian, you store it in Obsidian, and you deploy it to your agent operating system.
[4:38] And this is infinite, because most memory setups just work on what you feed the AI — it reads, and that stops improving the day you stop typing. The Infinite Context Engine is a loop. So every chat your agents have gets written back into Obsidian — automatically, one file per day. So the vault doesn't just store what you tell it; it stores what your agents do. Then the next agent reads that and it gets better and logs more. And then the vault gets richer — your work trains your vault, your vault trains your agents — they go round and round, smarter every time. As a loop.
[5:08] So this is a really powerful way to just dramatically improve your systems every single day. If we go into one of our agents and we ask it for personalized ideas — like, for example, Hermes — it's going to understand exactly what we've been working on recently, as you can see here, and it's going to have personalized recommendations based on what we're going to do, and that all comes from the memory system. Then that links to, for example, free Claude Code to call into Hermes, and everything else that we have inside this system.
[5:29] It's your whole mind as a galaxy. If you look at this system here, every memory is a star, every star links together. Mine has about 364 stars and 1,200 links right now. And every link between notes is a line of light. The stars you've touched most recently glow the brightest. It's not just pretty — it's actually how you find things. So if you need a memory from three weeks ago, you don't dig through folders; you search a galaxy or follow the links from one star to the next. Your agent is doing exactly the same thing.
[5:58] When they're seen, this is what organized really looks like. You might say, well, my notes are a mess, this won't help. But that's the point — the galaxy connects the mess for you. Wiki links pull related notes together so you can see the threads. Omi, your chats, your notes — and then that goes out to Hermes, Claude, Open Claude, and every other CLI they use.
[6:21] If you think about this: the old way, without a memory, is like you lose loads of times because you have to explain who you are every session. You pay the same context in every tool. You get generic answers that don't know your business. Each AI tool starts from scratch, and you forget what you decided last week. With the new way — with the Infinite Context Engine — it just knows you. Every agent wakes up already knowing you, one brain shared by a memory with Hermes. It gives answers built on your real goals and clients. It finds any past memory in seconds. The galaxy and the loop make this better every day.
[6:45] The other thing we built in over here is search. We can type in a memory like this, and it will search across all of our memories and pull up the latest notes on that. So you'll think, like, "Julian's meaning in life" — this is pretty amazing; it's actually understood what I see as meaning from life, and then it's detailed that into beautiful documentation.
[7:02] So that's how the whole system works — Obsidian, and then that links to all your AI agents. Really powerful stuff. You can see how it gets you back. You might say it's complicated — it's not, because it's really simple and easy to just implement. You might say this is technical — but you get your agents to do it for you. You might also say, well, does this use a lot of tokens? But we actually have token-minimization playbooks if you're worried about all that sort of stuff inside the app.
[7:25] So if you want to get the Infinite Context Engine built for you, you can wire this up yourself, or you can get the whole thing done inside the agent operating system — inside their platform, with the full Obsidian and memory-galaxy setup already wired, the Omi and Obsidian setup walkthrough, step-by-step coaching calls where I set up the memory with you, the community of 4000 operators running their six-hour stack, plus the prompts and a member map for your city.
[7:44] So if you want to get all of that, I've got a profit link in the comments description. Also get the full agent OS — if you want to get this, it's inside the classroom; go to "new daily updates" and you'll find the Agent OS over here, with the zip file to install it and a video tutorial. We also have daily tutorials on how the base actually works and what's useful. And inside the community, I answer all the questions, with a video tutorial every single day. Inside the calendar you can jump on coach calls, ask questions, wire this setup together, and meet up with cool people doing the same thing. Inside the map, you can meet people locally near you who are building with AI agents just like you.
[8:18] So feel free to get it — link in the comments description, or go to the app. Bye — so watch it.
---
NOTE: Light ASR (tiny) leaves a few ambiguous terms. Best-guess restorations:
- "blog.com" (end) → likely the AI Profit Boardroom / Agent OS app URL (unclear in audio).
- "Open Hall" / "Open Claude" → treated as Claude references.
- "room of 4000 operators" → "community of 4000 operators".
Raw file (verbatim ASR): ep_obsidian_omi.txt (same dir).
transcripts/ep_obsidian_omi.mp3
BIN
View File
Binary file not shown.
transcripts/ep_obsidian_omi.txt
+110
View File
@@ -0,0 +1,110 @@
# /home/vpsadmin/oracle/transcripts/ep_obsidian_omi.mp3
# m ...25 segments, 115s
...50 segments, 246s
...75 segments, 371s
...100 segments, 485s
DONE segments=105 -> /home/vpsadmin/oracle/transcripts/ep_obsidian_omi.txt
dian, this is a powerful way to have a memory system as you can see right here.
[12.5s] That just works and it has a powerful set of context and basically this trains all of our agents on exactly what we've been working on.
[19.8s] So that when we plug them all together, for example,
[22.1s] all of our own sounds, sounds chatchip, chatchipity, understand, to hermys.
[25.0s] All of our agents work together and they never forget context again.
[28.7s] Now at the same time, we also have notebook, a lem.
[31.7s] notebook a lem is one of the most powerful tools that I've ever seen for creating research.
[38.5s] And also the amazing thing about notebook a lem is like, you can generate, for example,
[43.0s] videos in one single click for free using this system as well.
[47.5s] So for example here, you can see all of these videos that we have built and these are free videos.
[53.0s] We can generate fully educational fully set up to convert fully automated in one single click
[59.5s] and we can use notebook a lem to do that.
[61.4s] Now when you combine the two, which is both for these free tools, the notebook a lem,
[66.2s] and then obsidian as well, you get a powerful second brain that can basically build an automate
[71.7s] anything together. So you might be wondering, okay, why would you want to use something like obsidian?
[76.1s] Well, for example, if I go into Claude here and I'm like, what did I work on yesterday?
[79.5s] It's not going to have much of an idea on what I'm doing, what I'm working on, what's useful,
[83.8s] what's not useful, etc. And so what you want to do instead is have all of your agent since I've
[88.3s] won't system as you can see where you've got Hermes, you've got Claude, you've got Antigravity,
[92.5s] you've got, for example, Codex and they all operate from the same shared memory. That means all of
[98.2s] your agents understand each other. You might say, why would you do that? Well, I call this the infinite
[102.6s] context engine. So it's one memory that every agent shares and your chat trains your vault,
[109.1s] your vault trains your agent and this is a loop that gets smarter forever. So all of your memories link
[114.6s] to, for example, all of your goals, your projects, your areas, your daily and a fin just links to
[119.5s] give a beautifully. And the problem with this, you know, if you use something like, for example, chatchipity,
[124.4s] is like, imagine if you hired a assistant every morning, they wake up, they remember nothing about you,
[129.3s] you explain your whole life again, your name, your business, your goals, your clients and
[133.0s] everything will day. Well, that's essentially what your agents are like right now because they don't
[137.6s] have a memory system, your AIS eGenius, but it has amnesia. Every chat you start from zero. The infinite
[143.1s] context, engine that I'm showing you today breaks this cycle for good. You might say, well, AIS memory
[148.4s] sounds complicated. Technical all this is, is a bunch of Markdown files organized into a nice picture.
[154.7s] That is literally it. This is just a bunch of files and folders that all link together naturally
[159.4s] to train your agents on all of your context. And you don't even need coding to do this. You can get
[165.4s] your agents to automatically update it for you and you get your agents to automatically come up with
[170.2s] the ideas for it. Now, there's three, three tools with one shared brand that I use to set this up. So
[175.4s] have OME, we have a sitopsidin, we have the AIS. OME is basically a system where we can, but automatically
[182.4s] record our screen, our microphone, it's looking at my device all the time. It's taking an example
[187.6s] and looking through all of what we're doing day today. And then you can see, for example, here,
[192.2s] it takes notes on me what I'm doing, what I'm working on, and it does this like every hour. Now,
[197.6s] when we're doing this, this actually exports into Obsidian. An Obsidian is where we have our knowledge
[204.6s] graph. We have everything linked together. We have all of our Markdown files. So what you end up with
[209.4s] is OME, creating the memories and taking the memories automatically on you. And then that plugs
[214.5s] into Obsidian. And Obsidian is where all your agents come together. So if we click on this, for example,
[220.2s] you can see that it has all of the information about this particular topic. And it's neatly organized.
[225.8s] It's formatted nicely. Everything links together. So we click on the app right before we can see the
[230.0s] links between that. And again, like this is just like kind of a Wikipedia, but few agents and everything
[235.6s] links together. And you've got to finish up one place. And so we've got OME taking the notes.
[240.4s] You've got Obsidian, gathering the notes that you've created and organizing them. And then that plugs into
[246.2s] an agent operating system, which is where you have a claw. You have OMEs. Every CLI all pointed at the
[250.6s] same volume. So every one of them or wakes up already known you of one memory in every agent. He might also
[255.7s] say, well, this works with one tool, but not all of mine. But the vault is just text files. And any AI is
[260.8s] trained to read a file path that can read your memory. So it's one set up called OMEs. Every tool
[266.2s] with the same brain working together. That's how it works. As a capture with OME, you organize,
[273.0s] we've Obsidian, you store it in Obsidian and you deploy it to your agent operating system. And this is
[278.6s] infinite because most memory setups just work on where you feed the AI, it reads and that stops
[284.0s] improving the day you stop typing. The infinite context engine is a loop. So every chat your agents
[289.0s] have gets written back in Obsidian. Or it's automatically one file per day. So the vault doesn't
[293.0s] just store what you tell it. It stores what your agents do. Then the next agent reads that and it gets better
[299.0s] and logs more. And then the vault gets richer and you work, trains your vault, you've all, you've all
[302.2s] trains your agents. They go round around, smarter, smarter every time. As a loop. So this is a really powerful way
[308.9s] to just dramatically improve your systems every single day. So if we go into one of our agents and we
[314.6s] ask it for personalized ideas like for example Hermes, what it's going to understand exactly what
[319.3s] we've been working on recently as you can see here. And it's going to have personalized recommendations
[324.6s] based on what we're going to do and that all comes from the memory system. Then that links to, for example,
[329.2s] free-clawed code to call it to Hermes. Everything else that we have inside this system. And so it's your whole
[334.0s] mind as a galaxy. If you look at this system here, every memory is a star, every star links together.
[339.7s] Mine has about 364 stars and 1,200 links right now. And every link between notes is a line of
[345.6s] light. The stars you've touched most recently glow the brightest. It's not just pretty. It's actually
[349.8s] how you find things. So if you need a memory from three weeks ago, you don't dig through folders. You
[354.0s] search a galaxy or follow the links from one star. The next, your agent, it's doing exactly the same thing.
[358.2s] When they are seen, this is what organized really looks like. You might also might say, well, my notes are
[362.1s] a mess. This won't help. But that's the point. The galaxy connects the mess for you. Wiky links
[366.5s] pull related notes together so that you can see the threads. You've got them. So I'm sitting in
[371.2s] works like so. Oh, me, your chats, your notes. And then that goes out to Hermes, Claude, Open Claude, and
[376.1s] every ever CLI they use. If you think about this, the old way without a memory is like your ludes, loads of
[381.2s] times because you have to explain who you are of a session. You pay the same context in every tool. You get
[386.2s] generic answers that I've already figured you're business, each AI tool starts from scratch. And you forget what
[390.2s] you decided on last week. With the new way, with the infinite context engine, it just knows you say every
[394.7s] region, wigs up already knowing you, one for shared by a memory with Hermes called Open Hall answers
[400.2s] built on your real goals and clients. It finds any past memory in seconds by the galaxy in the loop
[404.8s] makes this much everyday. The other thing that we built in over here is we can search. So we can type in a
[408.7s] memory like this and it will search across all of our memories and pull up the latest notes on that.
[413.1s] So you'll think like Julian's meaning in life. This is pretty amazing. It's actually understood
[417.9s] what I see as meaning from life. And then it's detailed that into a beautiful documentation.
[422.5s] So that's how the whole system works. Only obsidian and then that links to all your AI agents.
[427.7s] Really powerful stuff. You can see how it gets you back. You might say it's complicated. It's not because
[431.4s] it's really simple and easy to just implement. You might say, for example, this is technical. But you get your
[437.8s] agents to do it for you. You might also say, well, does this use a lot of tokens? But we actually have
[442.2s] token minimization playbooks if you worried about all that sort of stuff inside the app off and forth.
[445.8s] So if you want to get the infinite context engine built for you, you can wire this to give yourself with the
[451.0s] or you can get the whole thing done inside the agent operating system inside their
[454.8s] platform. With the only setup, the memory Galaxy F and L. So the full agent see it with the memory loop
[459.8s] and the Galaxy already wired, the only and obsidian setup walk through. So that by step, coaching
[464.0s] calls where I set up the memory with you, the room of 4000 operators running their six hours. Stack plus
[468.8s] the prompts the S. And a member map for your city. So if you want to get all of that, since I'd get
[472.6s] a profit, link in the comments description. Also get the full agente. If you want to get this,
[476.6s] it's inside the classroom and then go to new daily updates and you will find the agent OS over here,
[481.1s] with the zip file to install it in a video tutorial. We also had new daily tutorials, the base and we're
[485.4s] actually works and what's useful. And then inside the community, I answer all the questions inside here,
[489.7s] with a video tutorial every single day. Inside the calendar, you can jump a week on coach calls,
[494.0s] ask questions. You can wire this setup in together and you can meet up with a cool people doing the same thing.
[498.2s] Inside the map, you can meet people locally near you who are building with AI agents just like you.
[503.1s] So feel free to get it, link in the comments description or go to the app on
[505.9s] blog.com. Bye, so watch it.