4 ms·
Exactly. It's not appropriate for this type of question. LLMs can summarize the global internet sentiment about something pretty well, but they aren't the right
by queuebert 2y ago
Exactly. It's not appropriate for this type of question. LLMs can summarize the global internet sentiment about something pretty well, but they aren't the right architecture for answering precise scientific questions, nor will they ever be. It's like asking an old, experienced physicist to guess answers to things instead of solving equations.
- ben_w 2y agoThey're already better than I'd have expected possible, so I wouldn't say "never" — but also yes I agree with you, currently I'd compare their output to an off-the-cuff answer from an experienced professional (or an undergraduate trying hard).
- turzmo 2y agoAs a physicist I can say that unless ChatGPT is asked a definitional question, or a very common question (whose solution is likely described many places on the internet anyway), it is very likely to be wrong. Personally more than 90% of the questions I've asked it have been flat-out wrong so I stopped using it entirely.