'Stop Generation' does not work as intended; just stops streaming the output

[severe oyster]

Steps to reproduce:

1. Go to GPT4 (I believe this happens for GPT3.5 as well but it is fast enough to not be impacted as much, especially since the model has verbose outputs by default for my use case)
2. Give it a prompt, and stop the generation as soon as it begins
3. Try to do another prompt, or edit the original prompt and save it.

Expected result:

The generation that canceled should not be running in the background preventing you from queuing up your next generation; you should be able to start a new generation if you already stopped the last one.

Actual result:

Generic error stating that you're not allowed to generate more than one prompt at a time despite the fact you already stopped your previous prompt; this implies the generation still runs in the background.

Additional information

Browser: Google Chrome
OS: Windows 11

Misc. Context

I ruled out the possibility that it's me regenerating or prompting too quickly after I stop a generation; the amount of time required to do a new gen isn't changed in any way if you stop a generation, and it doesn't matter when you chose to stop it.
Therefore, the 'stop generation' option is pointless to use in the current state; it does nothing to improve your rapid drafting ability when you can remove a poor quality output by adjusting or regenerating your original prompt.

[severe oyster]

More evidence that it's not just me requesting too quickly; the entire generation still happens in the background and must finish before you're allowed to do anything else.

Evidence the same bug happens on GPT3.5

'Stop Generation' does not work as intended; just stops streaming the output

[cinder stream]

The key here is wait one minute. If you wait a minute, you are able to use it again.

It's a cooldown to avoid flooding the system.

[severe oyster]

My point was, if the response naturally finishes, you can already immediately respond afterwards. ‘Stop generation’ has no reason to not also allow for the same behavior, maybe with a slight cooldown of 5 seconds instead to make it easier to rapid draft programming solutions

[cinder stream]

It doesn't allow it, it's in the message wait one minute.

This would be less a bug and more a suggestion.

[plain thorn]

Why did you add music to a bug report?

Its funny but yeah

[severe oyster]

(was playing in the background)

i coulda muted but didnt bother

[severe oyster]

I figured it says 'or wait one minute' because that's the built in failsafe for whenever it somehow doesn't stop the generation.
In any case it's strange design to have 'stop generation' exist at all if it's gonna not actually stop generation but hide it from you permanently (even though it still finishes in the background), so not necessarily a bugfix but an oversight

Key words here are OR wait one minute. OR implies inherently it's the alternative, some sort of failsafe for extremely long generations and/or generations that didn't stop properly, because it being rate limited forever cause of a rare glitch would be terrible design obviously

[steel surge]

This happens for me too. Actually joined the server to report the bug. 😊

If you click "Stop", ChatGPT still thinks it's generating. It's not a 1 minute wait, it just depends on when the output would have been finished before.

So if the reply is 2 paragraphs, it'll be done in about 10 seconds. if it is 4 paragraphs, slightly longer.

[dull yoke]

YUP! Both the regular, free GPT and the Plus do the same thing in addition to a host of other bugs since the Aug 3rd update. As I've said before: We aren't the multibillion dollar companies OpenAi actually cares about- Shame isn't it.

[steel surge]

It's just odd that they haven't even noticed this

[severe oyster]

i think they maybe fixed this?