Events
CILVR Seminar: Research Questions, Directions and Thoughts in Large Language Model Reasoning
Speaker: Pavel Izmailov
Location:
60 Fifth Avenue, Room 7th floor open space
Videoconference link:
https://nyu.zoom.us/j/93891751343
Date: Wednesday, December 10, 2025
While the progress in AI has been staggering, our best models still have clear limitations. In this talk, I will argue that we still need new methods and ideas for training large reasoning models. I will describe some of the research directions I find promising, including questions in pre training, synthetic data, reinforcement learning, and generalization. I will also discuss weak-to-strong generalization in the context of AI safety and security, as well as developing capabilities.
Bio: I am an Assistant Professor at New York University and a Researcher at Anthropic. I am primarily interested in reasoning, reinforcement learning, language models, AI for science and AI alignment. Previously, I worked on reasoning and problem solving in language models at OpenAI. There, I contributed to the OpenAI o1 model and the work on weak-to-strong-generalization. I defended my PhD in Computer Science at NYU in 2023 under the supervision of Andrew Gordon Wilson.