I don’t understand why code interpreter cannot be used with disabled “chat history & training”. Is that really the case? I want to upload a file with confidential data to work with but am worried about it leaking out. Any advice please?
For whatever reason, it’s what OpenAI has decided. Maybe they really want the data to improve the model since it is still in Beta. Maybe it’s technically/legally difficult because they are spinning up a temporary environment to store your files. Maybe it’s an oversight. Hard to know since I don’t think they’ve documented it (yet).
My speculation is that it is a beta product, ergo anyone who uses it is a de facto beta tester, therefore they need to access the logs for training and evaluation to understand how the model is being used and how the model performs.
While OpenAI could be much more clear about the reasoning for not allowing beta features without chat history enabled, I’m 99% confident it comes down to the implicit social contract contract of giving up the data in exchange for early access to cool new toys.
Personally, I think it is a fair trade, but it certainly limits a person’s ability to use it with confidence for anything which requires confidentiality.