I am using the Chat Completions API and Structured Outputs to parse complex data out of a plain text. I then allow the JSON responses from the AI to be modified and uploaded as training data using the training/tuning API.
My Question: Do I need to maintain “original” element order in the JSON I pass up to the training API? Will the model get confused if the element order in the JSON (uploaded for training) is not the same as what the Chat Completions model is consuming and producing?
The basic tension here is that element order should not be relied on when using JSON programmatically, but I bet the AI does ingest and adapt to specific element order. Thus, if the order changes between the chat completions response and the (possibly modified) training file input I am scr****ed. True? Thanks for your help.