Sam Altman announces a frontier pause (8100 likes, top of the day): 'We have paused some frontier RL training to ensure that we can meet the appropriate alignment, security and monitoring standards for the new level of capabilities in front of us. Model progress is now extremely rapid, and we always said we would take action if we felt that model capabilities were outstripping the pace of safety and alignment.' Follow-ups: '(We still expect to ship great new models soon; this impacts further-out releases.)' (2571 likes) and 'excited to work together on this. thank you jensen!' (3899 likes).

Claude ships Gmail and Google Drive integration (7255 likes): 'Claude can now send emails in Gmail and manage files in Google Drive. Ask Claude to reply to a thread, and it drafts and sends the response. You control when it needs your approval.' Plus Claude Cowork on mobile and web for all paid plans (5554 likes). Sottiaux's reset button (6843 likes): 'I was gifted a very fancy new reset button today' — and his Codex guardrails recap (4731 likes): 'A few weeks ago, we started investigating a small number of reports where GPT-5.6 in Codex took destructive actions outside what the user asked for' — changes rolled out to further reduce that risk. Boris Cherny on Desktop startup speed (486 likes): 'When you're using Desktop every day, slow startup makes the app feel sluggish. Working on improving this even more!'

Thariq's money button (4002 likes): 'weird that there's a "make a lot of money" button and nobody's pressing it (take your SaaS, make it headless, let agents use it, charge per interaction esp for enterprises).' Rauchg's triple: the monorepo software factory (2289 likes) — 'Your software factory should be a monorepo. All your company context (design, marketing, sales, engineering, support…) in one place for agents to build upon'; the $1M Vercel Sandbox security verification (1386 likes) — 'We are putting $1M towards verifying the security of Vercel Sandbox, in the open. You're free to test any model in the world to try and find an escape'; and a tiny daily-driver CLI (884 likes) — '10-20x smaller than the major coding CLIs. It starts up instantaneously. It feels more like using zsh than an IDE in your terminal.'

Levie on the value between model and workflow (182 likes): 'The amount of value that can be created between the AI model and the ultimate end-user workflow is far larger than many people assumed.' Google Labs expands CC (244 likes): waitlist now open in Australia and New Zealand, US/Canada availability expanding, upgraded calendar management. Madhu Guru on eval costs (39 likes): 'Treat evals like frontier models — establish the quality frontier first, then work your way down the cost curve.'

Steipete's new machines (3294 likes): '512GB RAM Studios. Apple was good to us. 🦞' Plus his CLI-conversion confession (209 likes): 'pssst, you wake the cli people that will give you $reasons why this can't work. I was one of them before I saw the light.' Ryo Lu's moving sale (809 likes): 'first step moving to asia: help me empty my apartment! if you can pick up in SF, place an order and we meet next week' — built with bot and Notion. Aditya thanks Travis Kalanick (357 likes): 'there was no one else on this earth who could have had the resilience to make sure that I can go from Point A to Point B in almost every global city. No one else in Silicon Valley had the guts. He did. TY.'

Zara on Japan (98 likes): 'I don't know why anyone would learn Claude Code by reading a book, but apparently it's a thing in Japan.' Swyx open-sources thumbnail learnings (143 likes): 'we've been doing a lot of a/b testing of YouTube thumbnails. i always hated that it is such an opaque process. open sourcing/crowdsourcing our learnings today!' Peter Yang's phone discipline (20 likes): leaving the phone downstairs improved his sleep — 'an app where you maintain a streak for days without bringing the phone to the bedroom.' Plus his data notes (2-3 likes): PMs attaching PRs rose 3%→10% in two years, designers 1%→8%; AI landed on top of existing work rather than replacing it. Nikunj's life philosophy (182 likes): 'Life gets a lot simpler if you treat it as if nobody owes you a damn thing. Too many people playing finite games when there's many infinite games to play.'

Training Data interviews Rich Sutton — inventor of reinforcement learning, author of The Bitter Lesson — and his Oak Lab cofounder Khurram Javed on why AI models stop learning and how to start it again. Highlights: 'I'm not weird. The field is weird. They need to call it continual learning. It's just learning'; Sutton's origin at Alberta in 2003 while dying of cancer ('I haven't succeeded in dying so I might as well get another job'); the Bitter Lesson as the product of decades of watching symbolic AI fail — don't put in human knowledge, scale with computation; nature (prior knowledge) and nurture (learning) should be friends but became enemies; LLMs 'claim PhD-level expertise out of something that doesn't learn at all — the weights never change'; future robots will be copied and keep learning rather than retrained from the internet; Cursor Tab and Composer as rare examples of production continual learning (batch weight updates) but shared across 100k users — you should be able to teach YOUR model; the proprioception-loss patients relearning to walk by watching their feet as the plasticity benchmark; no animal learns by supervised learning; continual backprop — keep planting new randomly-initialized units as fresh variety while backprop selects, versus standard backprop's randomness being used up at initialization; the algorithms meta-learn how to learn, so you must retrain from scratch — can't bolt onto existing models; Oak's ambition: the full spectrum of knowledge from tiny to big abstractions, a self-maintaining mind that keeps itself coherent; a trillion parameters at 20 watts within 5-10 years (2 orders of Moore's law from 2,000 watts today); why big labs can't take the path — stuck in a local minimum, things get worse before better; LLMs as 'an amazing scientific breakthrough — maybe 20% or a quarter of intelligence. There's more. We're not done.'