Concrete open problems in mechanistic interpretability | Neel Nanda | EAG London 23

This talk is a whirlwind overview of several key areas of open problems in mechanistic interpretability, examples of work in that area, and how they fit into the bigger picture of the field. The talk is aimed at an audience without much knowledge of interpretability, and will briefly explain language model basics, but you will need some technical background to get the most out of this talk. This talk will be of most of interest to people who want to engage with the technical details of interpretability. Find out more about EA Global conferences at: https://www.eaglobal.org Learn more about effective altruism at: https://www.effectivealtruism.org