Hello, creators of CHAT-GPT or others users of CHAT-GPT,
I am a PLUS user and have been enjoying using the advanced product. However, I have encountered an issue that has been quite frustrating for me.
The new GPT-4 model has a limit of 100 messages for four hours, which I can accept. However, today I accidentally switched the model to GPT-3.5 after reaching the limit. There was no confirmation or response time, and as a result, all the progress and training I had made over the past two days has been lost. This has left me feeling very disappointed.
I am not sure if the model will be able to revert back to GPT-4 after four hours, but I am in urgent need of a solution. If the model can be restored to GPT-4 after the four-hour limit, I apologize for any inconvenience caused. If not, I would appreciate your technical support to help me switch back to GPT-4 and prevent all of my previous training from going to waste. Thank you.
Because of ChatML unfortunately I believe the only way to restore the conversation would be to repeat each message naturally. Can you not go back to the last GPT-4 message, and edit it to trim the “bad” messages?
I would actually be interested in seeing the difference between copy-pasting the whole conversation (as was possible before ChatML) and having the conversation naturally
Well,Let me wait for another four hours and see if I can reverse the previous conversation. If not, I will have to re-enter all of the messages one by one, which will be a huge task, especially with the message limit in place.
Finally,thank you very much for your response.
I think that’s correct.
One suggestion - I have experimented with Bardeen to harvest the chat sessions and store them in various things like Airtable or Google Sheets. One big advantage is the historical log of everything I’ve asked as well as responses; almost like Rewind. But here’s something exciting, Bardeen can reuse any preserved prompts and send them back into GPT with its own GPT integration playbook components. This makes all sorts of crazy workflows possible.
Amy conclusion here ? I have the same problem.
I haven’t found a particularly good method yet, so now I start to retrain. If there is a solution later, I will use the previous training results.
In addition, I have written to OPENAI officials, hoping that they can come up with a solution.
If I find a good way then, I’ll let you know.
Andersdyr via OpenAI API Community Forum <email@example.com> 于2023年3月18日周六 17:22写道：