Investigating RL for Interpretability with Caleb Biddulph (MATS, Google Gemini)

​We’re thrilled to host Caleb Biddulph, a MATS 8.1 scholar (mentored by Micah Carroll at OpenAI), former software engineer at Google Gemini, and founding president of Cornell Effective Altruism.

​Attendees will receive copies of MONA: Myopic Optimization with Non-myopic Approval Can Mitigate Multi-step Reward Hacking, a paper co-authored by Caleb. Following that, we’ll dive into a discussion about his current research on RL-driven prompt discovery and interpretability, as well as his insights on doing good technical research and pursuing careers in AI safety and alignment.

​Catered dinner will be provided—come for the food and great conversation!