Great talk, Deepak Nathani!

Congrats to the following paper authors attaining Outstanding Paper Awards at SEA Workshop! RPGBENCH: Evaluating Large Language Models as Role-Playing Game Engines Pengfei Yu, Dongming Shen, Silin Meng, Jaewon Lee, Weisu Yin, Andrea Yaoyun Cui, Zhenlin Xu, Yi Zhu, Xingjian Shi,

Congrats to the following paper authors attaining Outstanding Paper Awards at SEA Workshop! GEM: A Gym for Agentic LLMs Zichen Liu, Anya Sims, Keyu Duan, Changyu Chen, Haotian Xu, Simon Yu, Chenmien Tan, Shaopan Xiong, Weixun Wang, Bo Liu, Hao Zhu, Weiyan Shi, Diyi Yang, Wee

The best poster awards go to: 1. Go-Browse: Training Web Agents with Structured Exploration Apurva Gandhi, Graham Neubig 2. Scaling Open-Ended Reasoning to Predict the Future Nikhil Chandak, Shashwat Goel, Ameya Prabhu, Moritz Hardt, Jonas Geiping 🎉Congrats!

Kudos to all the speakers/panelists (Edward Grefenstette, Mike A. Merrill, Grégoire Mialon, Deepak Nathani @ NeurIPS 2025, Joseph Marino, Shuyan Zhou🛸NeurIPS, Qian Huang, Anthony G. Cohn, Eric Sommerlade, Fred Sala) and organizers (Guohao Li 🐫, Yuan He @ NeurIPS 2025, May Fung (hiring postdocs), Qingyun Wang, Fangru Lin @NeurIPS, Xingyue Huang @ NeurIPS 25, Alisia Lupidi,

The SEA Workshop at NeurIPS 2025 was a tremendous success, bringing together frontier discussions on building and scaling agent environments. We were fortunate to have outstanding participants, speakers and panelists (Edward Grefenstette Mike A. Merrill Grégoire Mialon Deepak Nathani @ NeurIPS 2025

ICYMI — the Terminal-Bench creators just laid out what actually matters for agent evaluation. Terminals > GUIs Containers for real rollouts TB 2.0 = harder tasks + deeper verification

Can AI self-improve on its own and reach superhuman performance? 🧠 In our Sima 2 paper, we dropped a Gemini agent into an unseen 3D world. The model acted as the task proposer, the agent, and the reward model - autonomously learning from self-generated experience. It surpassed

Great to see an AI lab doing and publishing science (as well as discussing engineering efficiencies)! Some of the other “frontier” labs should try it! Thx, DeepSeek!

Couldn’t agree more

Kimi.ai used policy mirror descent (PMD) for RL in Kimi k1.5/k2. Most take it simply as PG+KL with an updating anchor, but this is not the full story. Check our blog for some interesting findings about this algorithm: zhenghaoxu.notion.site/Revisiting-Kim…

x.com/i/article/2021…

It’s Chinese new year eve🤣.

Congrats for the milestone!

Thinking in the “right” amount - how to shape this reward signal is challenging.

Claude Code is evolving from “you’re absolutely right” to having real stance. Good code comes from arguments. Humans bring taste and system design; agents write and debug. There’s a growing illusion that code generation equals shipping. Without iteration, constraints, and