At Physical Intelligence (π) our mission is to bring general-purpose AI into the physical world. We're excited to show the first step towards this mission - our first generalist model π₀ 🧠 🤖 Paper, blog, uncut videos: physicalintelligence.company/blog/pi0

🚀 Ever wondered when RL agents can achieve constant regret? At #NeurIPS2024, we’ll present Cert-LSVI-UCB—the first algorithm with high-probability constant regret bounds for linear MDPs with minimal assumptions. 🌟 Looking forward to discussion at Vancouver󠁣󠁡󠁢󠁣! 🎉

Will be at #NeurIPS2024 next week (Dec 10th to 15th), happy to chat about (1) VLM, VLA, RLHF, etc; (2) PhD positions in my lab; (3) faculty positions at UNC. Feel free to DM/email me to chat if you're around!

Heading to #NeurIPS2024 tomorrow for presenting the constant regret RL paper and several workshop papers (SRLM, Hallucination in LLMs). Happy to chat about future collaborations / PhD / intern positions in my lab!

Our students Weixin @Kamome_233 and Hao-Lin Hao-Lun Hsu will be presenting this paper at #NeurIPS2024 today from 4:30pm to 7:30pm! Room: West Ballroom A-D Poster ID: 6501 Come to chat with us on how to generalize Thompson sampling to general reward function and noise setting.

Zhiyuan and I will be presenting this starting from 11AM today in West Ballroom A-D #6407 on the constant regrets in RL. Also looking forward to talking about any collaboration in the near future!

Congratulations to all the #UNCgrads! 🎓🎉 Find more information about Winter Commencement this Sunday at commencement.unc.edu.

Racism has no place in academia or society. Can’t believe this happened.

If you’re using the #ICML LaTeX template, there’s a typo in algorithmic.sty that prevents cross-referencing specific lines in the algorithm environment. The fix is simple: change \addtocounter{ALC@line}{1} to \refstepcounter{ALC@line} on Line 106. Credit: tex.stackexchange.com/questions/5234…

🚀 We introduce MJ-Bench-Video, a comprehensive fine-grained video preference benchmark, and MJ-Video, a powerful MoE-based fine-grained and multi-dimensional video reward model! 🎥🌟 🔍 MJ-Bench-Video evaluates 5 key aspects: 1️⃣ Text-Video Alignment – ensures precise adherence

Can we prompt robots, just like we prompt language models? With hierarchy of VLA models + LLM-generated data, robots can: - reason through long-horizon tasks - respond to variety of prompts - handle situated corrections Blog post & paper: pi.website/research/hirob…

Thank you AAAI for the honor & the fun ceremonies -- humbled to be inducted as a AAAI Fellow in esteemed company 🙏 PS. I am still around today in Philadelphia if anyone wants to meet up at #AAAI2025 :-) Thanks once again to everyone (students+postdocs+collaborators,

Huge thanks to my advisor Huaxiu Yao for guiding me through this paper, and big appreciation to my senior Peng (Richard) Xia for all the help! 🙏 #AI #DocQA #LLM #Agent

🚀 Launched MDocAgent—a multi-agent framework that crushes complex documents (text, images, tables!) with specialized AI agents. Outperforms SOTA by 12.1%! See details in Peng (Richard) Xia's thread.

We are going to present the following works at #ICLR2025 covering MLLM, Agent, RAG and Healthcare. Welcome to discuss with us!

I will travel to Singapore next week to present work on LLMs, alignment, and synthetic data generation at #ICLR2025. Feel free to reach out! 😁 Check the papers in the next post:

Flying to Vancouver tomorrow ✈️ to present our spotlight poster at ICML “Mitigating Object Hallucination in Large Vision-Language Models via Image-Grounded Guidance” — joint work with Linxi Zhao, Yihe Deng, and Quanquan Gu. Catch us over the poster session on Wednesday! Happy

Be sure to check out my (incoming) PhD student Shangzhe’s first work this Thursday afternoon at #icml25 Vancouver! I’m beyond excited to work with Shangzhe on more IL/RL projects ahead!

It’s happening now!