2 ms·
Unfortunately, that likely just doesn't exist. Everything suggests that these models are confident about their mistakes.
by cyanydeez 6mo ago
Unfortunately, that likely just doesn't exist. Everything suggests that these models are confident about their mistakes.
- feoren 6mo agoI mean, what I describe absolutely does exist, that's how LLMs work. The question is whether the relative weights are actually a good measure of confidence, and as the other reply to my comment points out, there are examples where it's not -- at least not the kind of "confidence" we really want.