4 ms·
Is it possible it’s learned to produce a correct-sounding answer to this specific question based on the fact that it was asked to chatgpt-3.5 previously? I don
by omginternets 4y ago
Is it possible it’s learned to produce a correct-sounding answer to this specific question based on the fact that it was asked to chatgpt-3.5 previously?
I don’t see why we should consider both to be blank-slate tests.
- simonw 4y agoIf you mean someone asks something of 3.5, and then a few moments later asks the question of 4.0 in a separate chat window, then no: it's absolutely not the case that conversations with 3.5 can impact conversations with 4.0 a few moments later. Every conversation with ChatGPT starts from a blank slate. You can prove this to yourself by learning how to interact with it via the API - and from OpenAI stating clearly that there is no memory outside of each individual chat session. If it DID have memory (and OpenID were lying about it) people would have figured that out already, due to the obvious and weird behavior that would result.