
Introduction to AI Control
AI Safety Fundamentals
00:00
Specific Failure Modes and Research Needs
The host lists potential failure modes like collusion and sandbagging and calls for more empirical testing of controls.
Play episode from 09:37
Transcript


