Yael Tauman Kalai

How to Avoid Debate: Scalable AI Safety via Doubly-Efficient Interactive Proofs

Liyan Chen, Yael Tauman Kalai, Zoe Xi

As AI models continue to develop powerful capabilities, it becomes critical that we are able to verify that their output is aligned with our intentions. A recent line of work focuses on verification via debate, a model of interactive proofs where two competing powerful provers, o…Read more