LLM Error Stubbornness

That's a OpenAi problem, not a LLM problem

Tip: Don't point out contradictions to a LLM AI. Just tell it to redo the wrong part. They can not figure out that they are wrong (yet.) They only spit out best effort attempts.

This is super interesting. Can you share some prompts that you recall had believable but wrong responses?

I find this behavior when using ChatGPT.If I ask for a task, and the output is not the one expected: I ask for the motivation that lead to the bad decisions. Then, ChatGPT proceeds to retry the task "incorporating" my feedback, not answering my question!!

I tried this and it seemed to break ChatGPT, it blurted out something which made no sense and then offered to regenerate it. How is it supposed to work?

It doesn't work this time because there are plenty of models, including GPT5 Thinking that can handle this correctly, and so it is clear this isn't a systemic issue that can't be trained out of them.

Thanks. I have also noticed this in some cases. Sometime GPT-3 does not follow the prompt. Please retry with a new query. It works in most cases. I will improve the prompt or to make sure this does not happen.

That's unfortunate. When an LLM makes a mistake it's very helpful to read the CoT and see what went wrong (input error/instruction error/random shit)

I haven't played with this model, but rarely do I find working w/ Claude or GPT-4 for that to be the case. If you say it's incorrect, it will give you another answer instead of insisting on correctness.

I work at OpenAI. This was unfortunately a bug, and we've fixed it since.