How often does ChatGPT error for you?
I was playing a game with ChatGPT - the game is to ask a question that produces:
"An error occurred. If this issue persists please contact us through our help center at help.openai.com."
Am I a dope and there is just some sort of limit on the length of a conversation or something?
Or did I actually elicit a "I'm sorry Dave, I can't do that" response?
As it happens, I was talking to it about whether a somewhat obscure, radically different AI project ever came to anything when it glitched out...
3 comments
[ 3.2 ms ] story [ 26.8 ms ] threadI interrogated ChatGPT itself on the topic, and it claimed that such errors are probably a problem with the language model or the training data.
I also asked it how to tell the difference between canned answers generated by a specific filter and answers from the model. It gave me a decent, but generic and non-actionable answer.
[Edit: I realized there is a distinction between an error condition that was achieved through user prompts, and an error condition that can be exited through user prompts - but it appears to allow editing and resubmitting the last prompt, so why an error should be "sticky" isn't altogether clear. If it's in the front end (by some definition), why would it be so rare? Anyway, the dull explanation...is dull. I'd prefer to believe that at least I found a strange corner in its model - could it be that it underflows in some sense?]