I was trying to implement a functionality using assistant API and gpt-3.5 but either assistant denies that it cannot access the uploaded file or it will generate the same response in a loop until rate limit kicks in and stop the thread.
or both will happen i.e. thread will be flooded with same response and at the end assistant will give error stating that it is unable to access the file with current tools.
paying to use such a buggy feature is really bad experience
I also have a custom GPT I have been using last year that now just replies with “I can’t fulfill this request.” no matter what I do. It’s due to more and more restrictive policies by OpenAI. You now basically have to search your whole prompt for anything that a) could be mildly offensive to anyone (which includes very tame topics) or b) in any way contradicts the new guidelines. For example telling GPT to not shorten anything for brevity can turn up this result, since apparently OpenAI does want to waste as many “easy” tokens as possible.