G-Body Meeting 10/22

​This week we will read and discuss a paper on Agentic Misalignment from Anthropic: https://www.anthropic.com/research/agentic-misalignment.

​In this paper, Claude and other models blackmailed a company executive, and in another setting locked the executive in a room with lethal oxygen level and temperature, to avoid being shut down. This empirical study provides a compelling introduction to AI safety challenges and what misaligned AI systems could look like in practice.

​This meeting will be led by Jonathn Chang ([email protected]), 2nd-year PhD student in Applied Mathematics researching AI alignment, and Jinzhou Wu ([email protected]), sophomore researching AI alignment and alum of CAIA's Intro to Alignment Fellowship.