Re:Are we in an AI Overhang?
by Smaug123 on Monday July 27, 2020 @02:19PM
(#23967865)
> it will just make up convincing rubbish
It's not certain that this is always the case. In at least one case I've seen, if you give it question-and-answer prompts where you don't demonstrate that you will accept the answer "your question is nonsense", it will indeed make things up; but if you include "your question is nonsense" as an acceptable answer in the sample prompts, then it will use it correctly. See https://twitter.com/nicklovescode/status/1284050958977130497 .
It seems that we have a lot to learn about how to use GPT-3 effectively!
It's not certain that this is always the case. In at least one case I've seen, if you give it question-and-answer prompts where you don't demonstrate that you will accept the answer "your question is nonsense", it will indeed make things up; but if you include "your question is nonsense" as an acceptable answer in the sample prompts, then it will use it correctly. See https://twitter.com/nicklovescode/status/1284050958977130497 .
It seems that we have a lot to learn about how to use GPT-3 effectively!