The Cursor launch post on Hacker News only had 14 votes and "I still can't figure out if this is just some sarcastic joke software" as the top comment. Incredible

Semantic search improves our agent's accuracy across all frontier models, especially in large codebases where grep alone falls short. Learn more about our results and how we trained an embedding model for retrieving code.

New eval! Code duels for LMs ⚔️ Current evals test LMs on *tasks*: "fix this bug," "write a test" But we code to achieve *goals*: maximize revenue, cut costs, win users Meet CodeClash: LMs compete via their codebases across multi-round tournaments to achieve high-level goals

Finding composer fantastic and small targeted tasks but ALSO really, really strong as a researcher because it is so fast and so well trained at reading code and docs intelligently! I've often found Codex struggling with an issue and I have it output the problem and then tell

But getting dressed in the morning, to which our civilization is now devoting so many of its marginal resources, is possessed of neither beauty nor symmetry. Indeed, putting one’s socks on before one’s shoes is quite different from putting on one’s shoes before one’s socks.

Think about this talk a lot. There was a time when people were bullish on "feed all the modalities to the LLM," but it didn't really pan out as I would have expected. The discrete / continuous divide remains a interesting challenge in deep learning.

defending today 🥲

Apparently there was already a discussion of this talk here after it was given live. But I missed it. So now will do: I do care about potential AI harms, but disagree with pretty much everything he says here! will elaborate in the response.

There was a lot of twitter conversation about this talk when it was given. Here is the full version. https://t.co/pOvn1YYHHR

COLM Keynote: Shirley Ho Building a Polymathic Foundation Model for Science https://t.co/APJTkvYJdZ

I found @cursor_ai's composer model extremely useful for many specific tasks. A very short feedback loop means I can prompt it 4-5 times, and it can instantly get to the code I wanted. Speed is really a 10x factor in AI

literally how do ML people survive without lifting tensor dimensions into the type system? isn't that like the number one thing youd immediately want

Correct take.

Hot take: DAgger (Ross 2011) should be the first paper you read to get into RL, instead of Sutton's book. Maybe also read scheduled sampling (Bengio 2015). And before RL, study supervised learning thoroughly.

ok composer-1 is pretty nuts, and the code it writes is quite nice. probably my new daily driver for many things not quite as galaxy-brain as codex, but it's SO fast that you can use it sync instead of async, and very quickly iterate on fixes. follows instructions very well

cursor just made every $200/month copilot subscription look like a scam dropped today with their own coding model what took 8 hours of manual coding now takes 30 seconds and it runs 8 versions of itself in parallel to pick the best solution while github's charging $20/month

David is an amazing PhD advisor and everyone should apply.

https://t.co/GbogvA48oR

Cursor just released a frontier coding model with 4x faster generation. They will be speaking at Ray Summit about their journey building a frontier coding model. - Training on 1000s of GPUs - Scaling 100,000s of sandboxed coding environments - Custom training infrastructure with

Overwhelmed by the number of Diffusion LLM papers? 🌊 Same here 😭 So I’m starting a Discrete Diffusion Reading Group (@diffusion_llms) with my favorite disciples @jdeschena and @zhihanyang_ ✨ We’ll cover everything—from theory to empirics, from language to molecules. Join