I'm experiencing inconsistent outputs from GPT-4 on Azure OpenAI, even with temperature set to 0 and no changes to my prompts or environment. Could these variations be attributed to unseen updates or bug fixes rolled out by OpenAI?

ChatGPT
, , , ,
1

If this is the case, how can developers effectively manage this lack of immutability and ensure consistent performance in production applications?