
AI Safety via Debate
AI Safety Fundamentals: Alignment
The Importance of Debate in Machine Learning
The simple version of debate discussed in section 2 does not capture many tasks we care about. There are several directions in which we can improve the model. A question may be too large to show to a human or to expect the human to comprehend. The next direction, answers may be too big. Similarly, the best answer to a question may be prohibitively large. To support large context, we let the agents reveal small parts of cue in their statements. This gives us hope that debate could resolve AI alignment without sacrificing model strength.
00:00
Transcript
Play full episode
Remember Everything You Learn from Podcasts
Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.