min

@sama not mutually exclusive. do some research.

i keep thinking i want the models to be cheaper/faster more than i want them to be smarter but it seems that just being smarter is still the most important thing

@_aidan_clark_ i feel like there is alpha in knowing such things

Absolutely crazy to me there is a whole generation of people who know what a transformer is but might not know what an RNN is (not meant to be a comment on Helen). Is this what getting old is like?

@avidseries high altitude environment and lifestyle not genes. anyone born there can do it

49 of the fastest 50 marathoners of all time had East African ancestry. Yes, it is in the genes.

Introducing Silico: the platform for building AI models with the precision of written software. Silico lets researchers and engineers see inside their models, debug failures, and intentionally design them from the ground up. Early access is open now. 🧵(1/10)

@jasminewsun just redistribute wealth based on how often LLMs use training data attributed to you.

Most people I know in AI think the median person is screwed, and they have no idea what to do about it. I spent the last 3 months talking to dozens of researchers, economists, and policy experts about AI's impact on work; including reps from every frontier lab and several Congressional offices. Unfortunately, I was not reassured. The AI industry is raising the alarm, but can't change course. These companies' core business model relies on the disruption they are warning about: their faith in full automation only makes them go faster. Policymakers are waking up, but still paralyzed by data and debates. Econ wonks disagree on plenty, but even the limited scenario looks like a "painful transition" that will disempower millions of workers. But an "underclass" is not inevitable, but rather a societal choice — and one we can and should stop. Instead of waiting for impact, we should start planning now to support workers through AI disruption. Whether policymakers can assuage concerns about economic security may determine if we get to reap AI's gains at all. New from me for @NYTOpinion. I put a ton into researching what I think may be the biggest topic of the year, so hope you read it (gift link here!) nytimes.com/2026/04/30/opi…

@GoodfireAI can I view exploding gradients with it?

@GoodfireAI insane

@zebriez anthropic

Inside Cursor ✅ Inside Notion ✅ Where should we go next?

@kukreja_abhinav it has worked in "small scale" with alpha zero. mad to think this will work on language given the action space

Pack it up if they get this to work

@thsottiaux @tmuxvim lol

@tmuxvim We actually have amazingly efficient models

btw, OpenAI is probably losing TONS of money on most codex users lol

@eliebakouch insane. like science from another planet

i might be very wrong here, but i don't think "no human data, no pre-training" is the right approach to get frontier models or scientific breakthroughs any time soon

@real_deep_ml lots of great people on paper, lots of undiscovered geniuses who get autorejected.

And it never was even a good system, a bot reads through a bunch of resumes that people tailored for bots Competing for a ML interview in a Kaggle style competition makes much more sense - you test people’s real skill at the job - you can’t use AI to cheat because everyone has AI - people don’t have to waste time making a resume they spend there time getting better

@gregisenberg i don't kmow what y'all be worried about. AI is gonna cure aging in like 10 years

From Sam Altman: “Adjusted for the subjective increase in how fast time passes, life is half over by 23 or 24. Don't waste time."

@sidtriv yes because the company is definitely just the code.

Diligence in 2026 is wild. My friends in PE are now spending the weekend before IC trying to rebuild the company they're acquiring in Claude Code. If the clone works, the deal dies. Cheapest moat test in human history.

@ankkala probably finetune mix didnt include lambda calc

This benchmark is telling a completely different story about the chinese models than the usual "kimi/deepseek is basically opus" What happened here?

@Brad08414464 probably would have been profitable enough to buy a serious lab outright

in an alternate timeline, Tesla + xAI would have been an AI infrastructure company like Coreweave, Crusoe, Nscale etc. this is more inline with Elon’s own strengths as an industrialist. he would’ve excelled at this type of business

@burkov for the data

Cursor is Elon's first purchase, which is a huge mistake. A coding agent harness is now open source (see Codex and Claude Code). The current design works virtually perfectly, so there's no need for a fundamentally new design. One can say that agentic coding is solved. One might argue that he bought Cursor for users, but users, as we know, switch between coding IDEs frictionlessly, so it's not like Twitter, where you cannot leave without losing followers. Cursor is an emperor with no clothes. A shell without substance. You only know it exists because it's been there before most others.

@icanvardar google invented modern ai.

google should just give up on ai at this point

@itsolelehmann this is why IPOs are coming soon. they are looking for someone to pay for their mistakes

seems like there an insane house of cards that could collapse if the economics change that much

how fucked are the frontier labs if chinese open source models come so close that for the average consumer/business it's not a big difference anymore? is this a real threat? or is this overblown?

@HarreBelly @chris_j_paxton its still a free supercar not just the absolute fastest.

@chris_j_paxton yeah not quite a lamborghini, but it's a free lamborghini

Deepseek v4 trained entirely on Huawei Ascend chips. A huge win for Huawei absolutely, and v4 looks like an amazing model. Not quite frontier of course, but its open weight