Post #1802974

2026-04-09 13:23 UTC

@pbloem @ngaylinn finally, I suspect the confusion is fuelled by the fact that there is a form of inference (deduction) that is 'truth preserving' but not only does that still require that premises are true, it's also very obviously the case that the overwhelming majority of inferences we conduct as humans are not deductive

Replies (1)

@ngaylinn@tech.lgbt 2026-04-09 15:22

@UlrikeHahn @pbloem I don't mean to imply LLMs are incapable of producing truthful answers. They usually do. My concern is that these algorithms come with no guarantees of accuracy, and their makers deny all accountability. Safe use of these models means remembering they are not oracles, even when they're presented as such. Neither are humans, of course. But we should not compare our tools to human beings!!! A tool for retrieving and manipulating facts should be held to higher standards. You could build such a tool using LLMs (we should!), but a standard LLM is not such a tool. They're "for entertainment only." If they have a "different mode" for factual information, then this is merely the LLM adopting an expert persona that produces more "truthy" output. But that doesn't mean the LLM is an expert, or that these outputs are more reliable, except in the empirical sense that they're wrong less often. Anyway, I think we're mostly agreeing at this point, just sorting out what we mean and want. 😜

Open ##1802975