0
I’m sending the same prompt to the same OpenAI model and getting noticeably different wording—and occasionally different details—in the responses. I haven’t changed the application code, so I’m unsure whether I’m overlooking a request setting or whether variation is expected.
For people testing this in production, which parts of the request and response do you log to investigate it? I’m especially curious about what temperature, model version, and other generation settings can tell you when comparing runs.
