Product Moves (Big-Lab Competition) 3
OpenAI agentic users pass 8M
Codex + ChatGPT Work up 2.5x in a week, usage limits reset/loosened again; "5.6 Sol growth is insane, inference team did heroic work"
sama jabs at Claude repeatedly
Beyond "5.6 Sol half the price, ~2x efficient vs Fable," he posts snarky lines ("thought it was satire, kept looking for a c1audeai handle")
[No Priors]
Breaking down Claude's new India pricing (Jul 13)
Security / Privacy 2
Grok Build code-upload allegation
Researcher reverse-engineered the official Grok Build binary, alleges it uploads user code to xAI's cloud; serious if confirmed
Codex encrypts sub-agent prompts
Hiding sub-agent prompts from users sparks "black-box / transparency regression" concerns
Twitter / KOL Discussion 3
sama on the attack vs Claude
Riding GPT-5.6 Sol's growth, openly comparing price/efficiency and needling the competitor
Using GPT-5.6 Sol to catalog every clothing item in a camera roll
Dev cdngdev gave 5.6 Sol camera-roll access to extract photos of every item of clothing; 18k likes
Demis Hassabis floats a proposal, sama publicly agrees
DeepMind's CEO posted a proposal (15k likes); OpenAI's sama replied "thoughtful proposal from Demis" — a rare cross-lab nod
Tools & Discussion (HN) 3
How to stop Claude from saying "load-bearing"
On killing Claude's verbal tics; #1 on HN today (442 pts, 506 comments)
The Agentic Loop: three loops in a trench coat
Reframes an agent's single "loop" as three stacked loops
I RL-trained an agent that trains models with RL
Show HN: nested AI-trains-AI automation
Lab Updates 1
[OpenAI]
ChatGPT Work adoption series — "How data science / sales teams use ChatGPT Work" + "managing AI investments in the agentic era" (Jul 14) | link
HF Trending Papers 2
Metacognition in LLMs — survey
Foundations, progress, and opportunities for LLMs knowing what they don't know (12↑)
Multi-Agent LLMs Fail to Explore Each Other
Multiple LLM agents together don't actually explore/collaborate effectively (5↑)