Being new to ChatGpt and to the ChatGPT Builder, I exceeded the 3 hour limit while my new custom chatgpt was condensing a video script for me. Finally I asked why it was taking so long and the response was generic as if we had never spoken before. After some looking around the previous 3 hours of fantastic training and customizing my chat appears to be lost. Does anyone know if it will show up tomorrow? Or do I have to try to start from zero?
1 Like
Hi!
Yes this happens more often than one would like to happen. I usually suggest to regenerate the first message that is out of context and try to see if the model catches back on again. If this doesn’t help, then a clean start is the best way to go.
If the transcription or condensing of the video failed than maybe you were just unlucky and can try again with more success.
You can also consider to copy and paste the old conversation into your new chat as new context. Sometimes it also helps to ask for a summary every so often and then keep this as a sort of back-up for when the model starts forgetting things (it’s running out of context length).
For starters this should get you going. I suggest you experiment with the few options the interface provides as they can be used in very clever ways to keep a good conversation going for longer.
1 Like
Thank you vb leader. I will work on your suggestions and hope to salvage what disappeared.
2 Likes
The instructions should still be in the settings for the custom GPT, so you should be able to replicate performance in a new thread.
1 Like
Yes, thank you Paul. They were there. I also found out that the training for the core operational function of the Custom GPT is applied by Chatgpt but there is no written record like there is with a record of a chat. I’m not sure I understand why, but that information doesn’t translate into anything I can save…so I am copying and saving my instructions to the Builder so I can restore it if my custom chatgpt suddenly forgets everything.
2 Likes
Great questions. To mitigate this I have built a simple versioning system for all my GPTs so that the Instructions, Knowledge, and Actions can always be restored with one click in case of accidental overwriting by the GPT Builder interface. This has helped a lot.
1 Like
Hi cass, that sounds interesting. I ended up redoing all the training and then copied and pasted those into word. But I can see this is going to be challenging to manage. Can you give me a brief idea how you create a “versioning system”? Is it just a matter of starting a version 1, 2,3 etc?