Re:LLMs can't self-correct in reasoning tasks, DeepMind stud...
by eichin on Monday October 09, 2023 @05:29PM
(#37825772)
Isn't this the essence of teddy-bear/rubber-duck debugging? You're forcing yourself to run your abstract concepts through the narrow channel of language and instantiating it in a way you haven't previously done (because if you had, you'd have solved the problem already and wouldn't need to be talking it through.) (I'm very specifically talking about the case where you explain and noone else contributes/responds, and you come up with the answer solely through trying to explain/speak it.)
[ Parent
| on HN ]
There are no comments on this comment yet.