How often does ChatGPT error for you?
3 comments
Irrelevant, All comments independent of the message fail and force you to restart the page, now at least those messages can be recovered and continue the conversation.
Sometimes restarting the page isn't enough. I forget exactly what I did, whether I logged out or whatever, but I know the third conversation when I got the error, the first level of restarting didn't revive it.
I 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?]
I 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?]
It also learns on its own quite fast. Last week it thought "Malta hetki", means "Moment of time in Malta". Now it knows better.
"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...