@hamishivi
Terminalmaxxing cannot be stopped 🤫
Introducing TMax v1.1! We improved our data and infrastructure, and managed to improve our models by up to 6 points on TB2.
TMax v1.1 dominates the Pareto curve for open terminal agents under 30B.
As usual, we release everything: our updated data, weights, infra, and training scripts are publically available!
Blog and resources👇View quoted post on X
This weeks paper club we have @hamishivi, first author of TMax speaking to us about the paper and building data recipes for terminal agents in general.View quoted post on X
RL envs are something that excites you do stop by for our paper-club! We only ask you to bring good questions 😉
luma.com/1j0drr5m
‼️The Bitter Lesson for context management: Giving LMs unrestricted control over their context beats human-designed SOTA!
Introducing 🩵Context Language Models (CLMs)🩵
- Natively manage their own context
- Treat context as a file
- Learn policies in CLM weights, no harnessView quoted post on X
Is RL actually making your LLM better?
Gains from RL are mostly on easy questions🤯
We're calling this the Matthew Effect for RL on LLMs. We then leverage async RL to solve harder problems by Never Giving Up!
paper arxiv.org/abs/2609.13443 blog mnoukhov.github.io/posts/ngu/ 🧵👇View quoted post on X
