Running into a weird bug where I start a new conversion with ChatGPT (chrome, desktop), using the GPT-4 model. I give it a python script, and ask it to do something. It replies, with purley an explination of the script. Then, when I reply once again asking it to do something, it asks me for the script again, even though I provided it in my first message. After telling (this is now my 3rd message to it) it I already provided it, it says my first message was X, which was not my first message, but my second.
#GPT-4 replies being sent as new conversations
rose garden
Seems like it stops processing anything after the last line of code (code is only 35 lines long)
GPT-3.5 doesn't have this issue, it reads the whole message
Code Interpreter model works perfectly (and actually solved the issue, compared to GPT-3.5)
regal tree
I am experiencing EXACTLY the same problem (The reason i did not update my plus subscription)
Here is a conversation example:
https://chat.openai.com/share/b57b2760-1b6a-473e-a0d7-063820aefdf0
Note that i broke the code in ttwo chunks to not exceed its token limitation.
But until yesterday i was able to paste the full code or multiple block as context,
Look at this chat about few days ago on the same project:
https://chat.openai.com/share/4efe2ffa-44ea-4125-bddf-ec4047f3e924
It had no problem "remembering " the previous code provided in early messages.
rose garden
Very weird. Hope they take a look and fix it