2026-02-28 00:07 UTC
@hen@social.lol I don't think it's a problem. To the extent that you got a useful response to a prompt is *also* the extent to which the answer was already in the training data.
If a statistical text-predictor predicts the most likely word to come next and that *happens* to be the correct answer, that means the answer is probably already out there at a statistically significant frequency.
1/2
Replies (1)
-
@michaelc@scholar.social 2026-02-28 00:09
@hen@social.lol Now, if you only tried an LLM and not search, then it's just bias. (You always find something in the last place you look because you stop looking when you find it.) BUT, if you couldn't find a solution via search first, that might have to do with search engines making search worse so that you're more likely to use LLMs. Still doesn't mean the LLM is "good" -- just that you're being manipulated into using them.