AI Safety Fundamentals: Alignment cover image

AI Safety via Debate

AI Safety Fundamentals: Alignment

CHAPTER

How to Play a Single Pixel Image Debate Game

With six pixels, debate with pre-commit turns a 59.4% accurate judge into an 88.9% accurate debate. The liar does much better if the liar can adapt in different parts of the game tree. Declaring the argument to be made in advance cuts down the space of arguments that must be covered. We have built a prototype website to play the single pixel image debate game with human debaters and judge.

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.
App store bannerPlay store banner