Poster un message

For generating completions of renowned poems, it is very tough to get GPT-3 to generate new versions except you actively edit the poem to force a change. Does it spit out completions that seem like it is pondering but it’s executing the improper algorithm, or it falls back again to copying parts of the enter ? I really don’t use logprobs a lot but I usually use them in 1 of three methods : I use them to see if the prompt ‘looks weird’ to GPT-3 to see wherever in a completion it ‘goes off the rails’ (suggesting the require for decreased temperatures/topp or increased BO) and to peek at (...)