4 ms·
Agree, "hallucination" as an argument to not use LLMs for curiosity and other non-important situations is starting to seem more and more like tech luddism, simi
by kristofferR 1y ago
Agree, "hallucination" as an argument to not use LLMs for curiosity and other non-important situations is starting to seem more and more like tech luddism, similar to the people who told you to not read Wikipedia 5+ years after the rest of us realized it is a really useful resource despite occasional inaccuracies.
- majormajor 1y agoFun thing about wikipedia is that if one person notices, they can correct it. [And someone's gonna bring up edit wars and blah blah blah disputed topics, but let's just focus on straightforward factual stuff here.] Meanwhile in LLM-land, if an expert five thousand miles a way asked the same question you did last month, and noticed an error... it ain't getting fixed. LLMs get RL'd into things that look plausible for out-of-distribution questions. Not things that are correct. Looking plausible but non-factual is in some ways more insidious than a stupid-looking hallucination.
- johnnyanmac 1y ago> to not use LLMs for curiosity and other non-important situations is starting to seem more and more like tech luddism We're on a topic talking about using an LLM to study. I don't particularly care if someone wants an AI boyfriend to whisper sweet nothings into their ear. I do care when people will claim to have AI doctors and lawyers.