EMNLP reviewer 2: why don't you compare with Llama 3.1? Llama 3.1 also has similar MCTS components for post-training.

Highest respect to the Japanese basketball team! They almost won France despite the huge physical disadvantage.

I love simple yet effective things. However, reviewers never agree with me on that.

Nice to see that NeurIPS this year implemented this. I have seen authors (from previous years) who literally sent daily reminders to the reviewers to ask for response.

Simple yet effective! Backstory: It was rejected by ACL23 due to its simplicity 🥲

Congrats xAI for getting stellar performances on MMLU-Pro and MMMU!

It seems that a lot of people don't see LLM as a part of NLP. They see it as a totally standalone interdisciplinary research area.

Want to know more about the recent progress of AI for Music? Check out our recent survey at huggingface.co/papers/2408.14…

Great effort led by Yujie, Bill and others in the team!

Maybe a dumb question. When I read the Chinchilla paper Figure2, I realize there is a bend (curvature) in the loss. However, when I tried to plot the curve using the fitted Chinchilla formula, there is no such bend. The bend itself also seems a bit weird to me because I have

After MMLU-Pro, we have MMMU-Pro. Great work led by Xiang Yue and other folks.