News for hackers. Stuff that matters.
Re:I disagree with Geoff Hinton regarding "glorified autocom... by zozbot234 on Saturday November 18, 2023 @02:11PM (#38323195)
The LLM equivalent to self-training is active learning. Roughly, this means come up with examples that the existing model is maximally uncertain about, and ask an outside evaluator which response is most correct. This maximizes the usefulness of outside help, similar to optimum experiment design in statistics.

There are no comments on this comment yet.