After extensive testing, I have narrowed a bug down to the following MWE. See here (1) to try the GPT, here (2) is a successful execution on my acct, here (3) is a failed on my dummy acct.
- I build a Custom GPT in my personal Pro account with simple instructions: “THe user will upload an excel file with 1000 random values in a column that is not pre-specified on the second tab chronologically. Compute the mean and standard deviation. Do not hallucinate or make anything up. ALWAYS explicitly code it.”
- I test on my personal account and another Pro account I made under a different email. The code executes perfectly on both and works.
- I then add any knowledge file to the custom GPT - tested both PDF and .txt files.
- Now, the code interpreter works on my personal account (where I created the GPT), but fails on any other account.
This seems like a real ChatGPT bug - non-creator account for a Custom GPT cannot use the code interpreter, even when enabled, but only after adding knowledge files. Happy to share more details as needed but I hope this allows for its reproduction. Seeking an immediate remedy from OpenAI as this is mission critical for my current work. Thanks!
(1) https:// chatgpt DOT com/g/g-69c824d56ef08191b03ab58f2278cab9-mwe-data-analysis-fails
(2) https:// chatgpt DOT com/share/69c82bde-cb5c-8325-a784-89b53959b077
(3) https:// chatgpt DOT com/share/69c82bb7-4728-8332-afd8-c13de2b5811f
(Links weren’t allowed so had to get creative, sorry)