3 ms·
Debate May Help AI Models Converge on Truth
- emptiestplace 2y agoJust wait until they gain meaningful context persistence and the consequent feedback loops.
- gildandstain 2y agoSounds like they applied insights from Mercier and Sperber's "The Enigma of Reason", namely that perceiving truth is hard, but sorting arguments by quality is much easier. Or -adversarial reason-giving as the engine of collective knowledge