On The Truthfulness of 'Surprisingly Likely' Responses of Large Language Models

Naman Goel

The principle of rewarding a crowd for surprisingly common answers has been used in the literature for designing a number of truthful information elicitation mechanisms. A related…