Events

Past Event

Machine Learning and AI Seminar Series: Jason Weston

December 12, 2025
11:00 AM - 12:00 PM
America/New_York
School of Social Work, 1255 Amsterdam Ave., New York, NY 10027 311/312

Speaker: Jason Weston, Research Scientist at Facebook, NY and a Visiting Research Professor at NYU

Registration for all CUID holders is preferred. If you do not have an active CUID, registration is required and is due at 12:00 PM the day prior to the seminar. Unfortunately, we cannot guarantee entrance to Columbia’s Morningside campus if you register following 12:00 PM the day prior to the seminar. Thank you for understanding!

REGISTER


Title: Self-Improvement of LLMs


Abstract: Classically, learning algorithms were designed to improve their performance by updating their parameters (weights), while keeping other components, such as the training data, loss function, and algorithm, fixed. We argue that fully intelligent systems will be able to self-improve across all aspects of their makeup. We describe recent methods that enable large language models (LLMs) to self-improve in various ways, increasing their performance on tasks relevant to human users. In particular, we describe methods whereby models are able to create their own training data (self-challenging), train on this data using themselves as their own reward model (self-rewarding), and train themselves to better provide their own rewards (meta-rewarding). We then discuss the future of self-improvement for AI and key challenges that remain unresolved.

Contact Information

Data Science Institute