Our first cohort of KRANIUM Harvard University undergrads is here! Learn more about this intensive summer research program, the students, their mentors, and their projects: bit.ly/3VPXzDd #AI #NeuroAI #ML

⭐New Paper Alert ⭐ How can your #RL agent quickly adapt to new distribution shifts ? And without ANY tuning?🤔 We suggest you get on the Fast TRAC🏎️💨, our new Parameter-free Optimizer that surprisingly works. Why? Website:computationalrobotics.seas.harvard.edu/TRAC/ 1/🧵

DPO is extremely effective when scaled with on-policy data scored by synthetic judges. In this blog post, we conduct a case study on summarization, directly optimizing for objective metrics like Q&A accuracy + compression instead of typical subjective ones.

What's missing from #LLM #chatbots? Read thoughts from #KempnerInstitute's graduate fellow Kenneth Li in this The Gradient blog:

So excited to have my first #NeurIPS paper!! And so very lucky to have the best collaborators and mentors for this project.

Speed up your DPO training with zero compromises using ✨prefix sharing✨! Prefix sharing speeds up training by 1.2-1.6x 🚀across the datasets we tested and is numerically identical to standard DPO training. Paper: arxiv.org/abs/2410.20305 Code: github.com/frankxwang/dpo…

I'm visiting exceptional startups in New York this week on the Harvard Startup Trek organized by my brilliant friends Kelsey Chen, Jonathan Lei, and Eva Tuecke. Today, we... – Visited Embedded Intelligence (ei.inc) and Medal, thank you Pim de Witte and team

Charts and graphs help people analyze data, but can they also help AI? In a new paper, we provide initial evidence that it does! GPT 4.1 and Claude 3.5 describe three synthetic datasets more precisely and accurately when raw data is accompanied by a scatter plot. Read more in🧵!