Re:Legal models hallucinate in 1 out of 6 (or more) benchmar...
by rustcleaner on Friday May 31, 2024 @06:52PM
(#40541166)
My sense is an LLM is like Broca's area. It might not reason well, but it'll make good sounding bullshit. What's missing are other systems to put boundaries and tests on this component. We do the same thing too: hallucinate up ideas reliably, calling it remembering, and we do one additional thing: we (or at least the rational) have a truth-testing loop. People forget that people are not actually rational, only their models of people are.
[ Parent
| on HN ]
There are no comments on this comment yet.