Guests Richard Ngo, Lawrence Chan, and Sören Mindermann discuss the dangers of artificial general intelligence pursuing undesirable goals. They explore topics such as reward hacking, situational awareness in policies, internally represented goals in deep learning models, the inner alignment problem, deceptive alignment in AI systems, and the risks of AGIs gaining power. They highlight the need for preventative measures to ensure human control over AGI.

Lawrence Chan discusses using mechanistic interpretability to create compact proofs of model performance. Topics include exploring proof strategies for small transformers, the importance of mechanistic understanding for tighter bounds, challenges in scaling proofs, and addressing structuralist noise in model behavior.