I have tried 3.5, 4, and Bard, all are pretty bad at Go. The hallucinations are quite troublesome in all cases. They create unused vars and then use vars that aren't defined. They all seem to have trouble with this. Even in Python, I find them importing and then never using something.
I found the code more reliable if I kept the function as atomic and low level as possible and to start a new chat if it got off the rails and loopy. I'd start with a better and more conscience queery to get a mlre thoughtful answer.
Could be, but I've been using 4 up until tonight without a single problem. 4 was magic until this evening, TBH.
Maybe if it continues I'll do as you suggest.
Interesting, please keep up abreast of your encounters with the good GPT.
Do you have API access? If so, are you able to find a prompt/response pair where the model `gpt-4-0314` (supposedly the original publicly released version) does worse than `gpt-4` (the current model)? If you don't have API access, you can share some transcripts where the latest version of `gpt-4` seems to have performed worse than it did in the past, and I will try them myself against `gpt-4-0314` and post the results.