AI Safety Thursday: The Limitations of Reinforcement Learning for LLMs in Achieving AI for Science

​What are the fundamental bottlenecks in reinforcement learning for scientific AI?

​LLMs combined with Reinforcement Learning (RL) have unlocked impressive new capabilities. But do we simply need more scaling to reach the next step: AI for science and research? If not, what are the limitations, and what else is required?

​In this talk, Yongjin Yang will share research on three fundamental bottlenecks of reinforcement learning for LLMs: skewed queries, limited exploration, and sparse reward signals. We'll also discuss potential solutions, and safety concerns from advancing RL toward AI for science.

​Event Schedule
6:00 to 6:30 - Food and introductions
6:30 to 7:30 - Presentation and Q&A
7:30 to 9:00 - Open Discussions

​If you can't attend in person, join our live stream starting at 6:30 pm via this link.

​This is part of our weekly AI Safety Thursdays series. Join us in examining questions like: 

• ​How do we ensure AI systems are aligned with human interests? 

• ​How do we measure and mitigate potential risks from advanced AI systems? 

• ​What does safer AI development look like?