Assistant has access to files from other assistants and threads

Here I show “another company’s documentation”

Except it is not. Depending on how you look at it, it is either fabrication and hallucination, or inventive inference. In any case, it is text prediction from a model not trained to deny unless you enhance that denial.

Is it reciting somebody else’s files? No, just crafting you an age discrimination lawsuit on demand.

Screenshots take from OpenAI’s prompts playground, using the Chat Completions API. You can have a fine-tuned company assistant with no actual model tuning also!