Detecting hallucinations in large language models using semantic entropy

Re:Detecting hallucinations in large language models using s... by Def_Os on Sunday June 23, 2024 @11:49PM (#40772524)

This is a good point. If you're worried about factuality, entropy is generally bad. But creative uses might thrive on it.

[ Parent | on HN ]

Re:Detecting hallucinations in large language models using s... by surfingdino on Monday June 24, 2024 @02:16AM (#40773013)

https://www.storycubes.com/en/ have been around for a while and do not require a huge data centre to create random ideas.

[ Parent | on HN ]

Re:Detecting hallucinations in large language models using s... by tliltocatl on Monday June 24, 2024 @02:47PM (#40779377)

But if you care about factuality, why would you use generative at all, rather than RAG or some old-school fuzzy full-text search? The whole things sounds like "we have a technique (LLM) that gives results once considered impossible, so it must be the magic box that solves all end every problem".

[ Parent | on HN ]