Sharing a link creates conversations that cannot be loaded or deleted.

[smoky shell]

I have a growing list of Unable to load conversation errors and there is no way to delete them.

1. Create a new conversation
2. Share the conversation (a great baseline prompt tool)
3. Load that conversation via the link
4. Continue conversation (using the baseline prompt)
5. Revisit the conversation later from the history

ERROR Unable to load conversation
a557dfb5-d6b4-4e26-a5b9-19b19d2a4cce
043c9158-e648-4bbf-a280-e792ff2ca2ab

6. Attempt to delete conversation

CANNOT BE DELETED

Any ideas how to remove these orphans?

[sacred stag]

Bumping this, as I too have a bunch of glitched conversations created by this issue that I would like to delete as well.

I have ways to prevent it from happening, but the bugged conversations remain and are quite annoying.

[crimson tendon]

I have further data and reproducibility steps to add to this. Once the conversation 'breaks', clicking on it again, after having left it, results in this message, top center of the screen:

Rapid clicking gets several of them. Briefly one can attempt to click the trash can icon, however, it disappears even if one manages to open it - and an additional 'unable to load' warning appears.

I can reliably 'bug out' one of these conversations.

To do so:

Generate a conversation.

Share it.

Open the shared link.

Continue the conversation.

At this time, the url shows something like:

https://chat.openai.com/share/42ebc782-3e7f-4251-ba19-b305496110ae/continue

To most reliably trigger this bug, edit a previous user message while the url still shows that /continue. The url will change, either immediately or eventually as the conversation is continued normally, but at this point there's no record of the conversation in the user's list of chats.

Note where I have selected to edit:

This is before I added, one input at a time, a duck, a squirrel, and a dog to the tests.

At this 'before point' the AI should not know about those test additions. When I edit to ask what has been tested, the AI should tell me nothing specific, just that we are testing.

However, instead:

You can see my edit branch.

You can see that the model 'knows' about the conversation past the edit point.

And what doesn't show in my screenshot, with my edit the conversation has been added to my new messages list. It was also given its unique url.

if I continue the conversation now, it's fine and safe, I can.

And it continues to 'know' what is in the other branch.

But when I go back to that first edit point and attempt to input there:

I initially get this error message:

If I click regenerate I get this error message: