Assistants not retrieving given data properly with code_interpreter

API
1

I have noticed that when I have a json file and code_interprenter only turned on you can ask some basic questions and it says the randomest things.

Example I have a json file that says:
eductation: “computer science at MIT”

Assistant response to “Where did he study?”

  • African studies and economics in UK

like not even relevant at all

2

Is the assistant opening the file and calling code interpreter? If not, the model will hallucinate a response. You can try prompting it to say something like “Open the file and answer the question: where did he study?”

2 Likes