CAIA Speaker: Leo Gao (OpenAI)

​Who: Leo Gao (IN PERSON), OpenAI
When: January 31, 2-3 pm PT
Where: Broad 100
Zoom: https://rit.zoom.us/j/99442271790

​Talk: An Ambitious Vision for Interpretability
Interpretability often settles for partial explanations. In this talk, Leo will argue that it is still worth aiming for the ambitious goal of fully understanding neural networks. He will share recent work on circuit sparsity, explain why it is an important step toward ambitious mechanistic interpretability, and outline the most promising next directions for the field.

​About the speaker: Leo joined OpenAI and has worked on alignment research there for over 4 years. His work spans interpretability, generalization, chain-of-thought monitoring, and overoptimization. Previously, he co-founded EleutherAI and helped create The Pile and GPT-Neo.

​After the talk:
If you are interested, we will host an informal social with Leo from 3 to 5 pm in Broad 100. You will also have the option to sign up for 1:1s with the speaker.

​Everyone is welcome: No specific technical background is required. Come learn and ask questions.
And yes, we will have pizza and boba.