Mid Chat Switch Model

[robust crag]

#Bug Report:

Steps to reproduce:

1. Start a voice chat in standard Model 4
2. Continue to chat as normal

Expected result:

A voice chat using standard Model 4

Actual result:

Chatgpt Model Spontaneously switches from 4 to 3.5

Additional information

Browser: IPAD App
OS:``` IOS latest update ipad pro 2022

[fallow monolith]

Did you perhaps run into your 50msgs/3hrs GPT-4 usage cap? When that happens, you're switched over to 3.5, and unfortunately there's no way currently to switch back.

[robust crag]

Thanks for your reply, that would certainly explain it. This has never happened before, this conversation has been going on for hundreds of messages regulary meeting the cap. It is a deep and rich creative chat that I require the 4 model for. This is a very disappointing and unhelpful occurance because I have trained this chat to be able to assist me in building my brands. Now I have start a new 4 chat and teach it all over again only for it to switch again? If its going to switch that should either be user controlled or just not do it at all. I already have a 3 chat trained on this subject for when I run out of messages.

[robust crag]

NB: this switch happens at the same time the automatic language detection model decides I am no longer speaking english (I am) and decides to translate what I am saying into Welsh and reply in Welsh.

[fallow monolith]

For what it's worth, if the chat had been going for hundreds of messages, the model had probably already forgotten a good amount of the older ones just due to context limits.

Perhaps you could start a new GPT-4 chat, and tell it you want to share some messages from an older chat to prime its context in this new chat, then go back and copy/paste some select context from the old chat into the new one? You might be able to get back to a working system quicker than you'd think!

[robust crag]

Thanks thats what I thought would be my only option, I just like the feeling that Chatgpt understands me, Ive invested alot of time in the conversation and it sucks to start again. Now I know there are context limits I wont be so concerned.