Agree with pretty much everything in this whole thread. I’ve been using GPT-4 again. I’m telling people that ask me 4o hallucinates too much and is unable to control the relative length of its replies.
I can hand it valid code, have it give me a bad example. Point out the error to it, give it correct code again, have it repeat the error… Then when asked it can correctly identify the error, following which it proceeds to produce the error again…
Lol. It cannot be trusted for even one-off prompts let alone as part of an automation or application.