News for hackers. Stuff that matters.
Re:Double Descent in Human Learning by malux85 on Wednesday April 26, 2023 @03:01AM (#35710368)
Overfitting is lack of genralisation, LLMs can produce novel output by using abstraction composition, which you cannot do without generalisation.

Other signs of non-overfitting: Abstraction laddering, task decomposition, novel (i.e. unseen) joke explanation

There are no comments on this comment yet.