Very excited to finally release our paper for OpenThoughts! After DataComp and DCLM, this is the third large open dataset my group has been building in collaboration with the DataComp community. This time, the focus is on post-training, specifically reasoning data.

it is not only about the candidate. ML internships at these institutions typically works like this: Prof pays 10-18k INR to a 3rd year student for data annotation task, where the project is well funded — instead of guiding them to produce something good. so - do not waste

super cool people in batch 5 of Lossfunk residency 🫶 looking forward to learn from you all ❤️

It is super fascinating to see that open ai is still not being able to beat humans in coding competitions. i mean i feels absurd, this much of money, compute and the knowledge of complete world still you are not able to beat human programmer. i think OpenAI is ngmi. all bets

this is super cool 👀🫶

We are starting a journal club on computational chemistry :) lets discuss cool things and learn together. this was my friend's Ranjan Bassi . would love to meet all the comp chem people here. catastropiyush.github.io/journal_club_c…

Open Thoughts delivers again. Congrats team for a small but powerful reasoning model. Writeup: open-thoughts.ai/blog/ot3_small

Congrats to our post training team who worked on the Hermes 3's dataset - Teknium (e/λ), NobodyExistsOnTheInternet, and outside contributor interstellarninja - on creating the now #1 Trending dataset on HuggingFace!

super cool work, congratulations Sachin 🔥

Want to build your next agent? I published what I learned recently about what works and what doesn’t.

We have Physics Wallah jumping into AI race before GTA 6 🙌

I hope you'll love what's dropping tomorrow! Anindya and Cory Kornowicz have been hard at work and we can't wait for you to try it!

The first AI-powered structure prediction editor, powered by Boltz-2 with bulk structure prediction, is now available on the LiteFold Platform. Links in comments. Exciting updates coming soon! 🚀

Too late to join the party? No internet/ gpt/ claude - just raw curated thoughts and yc application 👀 Y Combinator

We published notes on a new exciting paper that's dropped recently 👇

🫡

Giving a talk on RL and worldmodels for robotics today Lossfunk. Show up. lu.ma/55ymsyg8

super amazing work — third wave of RFT, we will be looking at more and more realistic simulations to train agents performing complex tasks.

More money -> more devs -> more experiences -> more users -> more monetisation -> more money -> more devs..

AI Agents are terrible at long-horizon tasks. Even the new GPT-5 model struggles with long-horizon tasks. This is one of the most pressing challenges when building AI agents. Pay attention, AI devs! This is a neat paper that went largely unnoticed. Here are my notes: