
27 - AI Control with Buck Shlegeris and Ryan Greenblatt
AXRP - the AI X-risk Research Podcast
00:00
Ensuring Safety Through API Usage Monitoring and Invariant Maintenance
This chapter emphasizes the need to closely monitor API usages in models to ensure safety and prevent high-stakes failures. It discusses the importance of monitoring internal usage of APIs, maintaining invariants to prevent failures, and monitoring a subset of requests at the right frequency.
Transcript
Play full episode