Altman's double: the framework post (7781 likes): 'The world deserves confidence that American companies developing increasingly capable AI will act responsibly, especially as the trajectory of progress has steepened. Every frontier lab must deliver on this, and there is no reason any of us should come to work if we cannot. We welcome a federal framework that sets consistent safety requirements for frontier AI. But we do not believe we need to wait for an anti-trust exemption or legislation to begin'; and the two failure modes (7540): 'First, we could lose control of the future to AI. This is unacceptable; we are unapologetically on Team Humanity… Second, we could end up in a world with too much concentration of power.' Levie's pacing defense (168): '"Pacing" can be somewhat of a trigger word because it sounds like an arbitrary slow down of capability or a way to hobble competitors through undue regulation. However, the specific improvement goals laid out in Dario's are absolute necessities in AI development. We expect equivalent in areas like aerospace, life sciences, health care.'

The rest: Shipper's deadpan (285): 'started using Astra medium for simple tasks. call that pacing the frontier'; Peter Yang's perspective chart (44): 'While we're all debating whether AI will lead to a bad outcome for humanity, it's important to remember that most of the world hasn't even started using it yet. Pedro, Brex's CEO, likes to refer to a specific chart: Each dot is 3.2 million people. Every gray box is someone that never used AI in the world (84%). Green boxes are free chatbot users (16%). Orange boxes are people who pay $20 a month for AI (~0.3%)' — plus 'I'm really curious who's actually managing this account' (39); and Aditya's one-liner (12): 'Imagine a world where we didn't have to rely on any one organization to be good and virtuous.'

Cherny's flex (1937 likes): 'Fable solved the Cyphral Distich (a 370 year old cypher). Super cool way to use Claude.' Rauchg's zen (2523): 'You don't get it, I build software to relax.' Steipete's engineering trio: the perf win (643): 'Next release (or dev channel) does worktrees ~80% faster via apfs/btrfs/xfs/ReFS folder clones. Also saves lots of disk space'; the Rust truth (102): 'This is all written in Rust. Because you can't escape Rust'; and the upstream note (45): 'I'll test this for a few weeks and if we see that this helps the majority of users I'll try to get this into codex!'

Amjad's pair: the pricing relief (179): 'It was painful to see so many users being priced out of AI coding for a while. But now it's free to build again!'; and 'Nominal determinism strikes again' (126). Plus Nan Yu's zinger (13): '37 signals would have called the the Nuse Feed' and Madhu's Muse note (6): 'Stud traders like Kaushik love Muse!'

Zara's UX gripe (170 likes): 'Anyone else notices this problem with Astra? Astra: I have done X. Me: This is wrong, you should do Y instead. Astra: You're right. I should do Y. Me: … Why doesn't it just go and do Y, like the other models??' Turck's meme (333): 'Seed investor when a company he forgot about years ago gets acquired out of nowhere.' Nikunj's trio: the valuation bug (75): 'Applicants: your entry valuation being higher and your total comp being higher is often a bug and not a feature… Just because a company has raised at a higher valuation doesn't mean they are realistically more secure' — plus the mirror meme (35) and the checklist (0): 'All the questions to ask when you're joining a startup.' Garry Tan's teaser (219): 'The content about startups we do is merely the teaser for the specific advice we give 1:1 when you actually are a YC startup.' And Peter Yang's bug report (8): 'Record & Replay seems broken and won't trigger.'

A brand-new No Priors with Rene Haas, CEO of Arm and head of SoftBank Group International. Core threads: the opening thesis — 'There's no computing problem that's ever been invented that doesn't utilize and can't utilize the microprocessor. It is the heart of everything'; Arm's two positions — IP licensing into phones/data centers/autos (which means 'we kind of see everything' across the supply chain) plus its first physical product (the chip unveiled at Hot Chips), which put Arm 'up to our waist' buying substrates, wafers, and memory itself; why physical chips happened — the evolution from components to compute subsystems (the Lego/blueprint analogy; demand was 'insane') to product, with Meta as the first domino: 'They wanted a general purpose agentic CPU. There wasn't anybody out who could give it to them'; surprisingly little ecosystem pushback — more software in the wild benefits everyone, and Jensen, Ronnie, Ahman, and James Hamilton all congratulated the launch; the 98.5%-gross-margin IP business vs. the new supply-chain muscle (fabless, but needing TSMC/Samsung wafers, Micron/SK hynix memory, backend/layout/bring-up teams); AI inside Arm — 80-90% of engineers use it daily, mostly on verification ('The largest amount of time is in the verification, the validation, the debug' — 'If we were to shut it off, it's like being in the 1990s, you've got internet and you're saying only internet between the hours of two and four… It'd be anarchy. The genie's out of the bottle'); still immature: RTL generation and physical design, because models train on public data and chip knowledge is proprietary — but Arm's documentation/test benches are a built-in advantage ('If it's unusable and untestable, it's actually untrainable'); in 5+ years, idea → GDS2 for straightforward designs is 'quite possible'; supply chain: constrained for 3-5 years at least 'so long as the transformer is the unit of energy'; the next bottleneck is data center buildout itself (labor, local restrictions); on the AI bubble: 'Are we oversupplied to demand? Not even close'; SoftBank as a single dominant shareholder is an advantage — plus SoftBank Neo, the intent to become a neocloud that can be a home for young chip companies; Haas also leads Ampere, Graphcore, and Stack AV ('I'm kind of in the room on a lot of discussions that Masa is having'); robotics: robotics 1.0 was purpose-built with high barriers, but learning robots + general-purpose mechanics + falling costs = a Jetsons future ('I don't think anyone's gonna have any interest in watching a sports league of robots'; the real utility is replacing human labor) — both humanoid and specialized form factors, and 'The robotic industry will be powered by ARM' (the brains already run on Arm); early adopters: factory automation, distribution centers, delivery ('a truck that has an autonomous [driver] is a robot of sorts'); policy: more US fabs are critical for national security (a Sematec fan from the 1980s Japan-Inc era), export controls are 'an infinite game' with no winner, but being the laggard means 'having the entire script dictated to you'; the data-center backlash is fear-driven (the electricians' union literally asked not to ban them; AI is 'the boogeyman for a lot of things'), and 'there is no downside from being the leader'; and the CPU thesis: training → inference shifts make orchestration the job — 'The token factory just generates all these tokens. It's like literally where are the trucks that are gonna take the tokens away and give them to the users? That's what CPUs do.' Edge AI is Arm's sweet spot: 'You just can't put a 50 watt GPU on your head.'