AI Ethics: Foundational Challenges in Assuring Safety and Alignment of Large Language Models

Abstract:

The talk will present the upcoming agenda paper on Foundational Challenges in Assuring Safety and Alignment of Large Language Models that Usman Anwar led with 35+ co-authors and advisors from NLP (Danqi Chen, He He, Yejin Choi), ML (Jakob Foerster, Florian Tramer, Samuel Albanie), AI Safety (David Krueger, Yoshua Bengio) and AI Ethics (Tegan Maharaj, Atoosa Kasirzadeh).

The agenda is very long, and the talk will begin by motivating why it is worth reading and then go on to selectively discuss some of the foundational challenges/topics.

Bio:

Usman Anwar is a second-year PhD student at University of Cambridge, advised by David Krueger, primarily working on AI Safety and deep learning. He is the recipient of Open Phil AI Fellowship and Vitalik Buterin fellowship on AI Safety from Future of Life Institute. He is primarily interested in foundational research on AI alignment, with a current focus on understanding in-context learning, and on understanding generalization behaviors of RL trained AI agents.