Got it. I’ll dig into whether others have encountered similar issues with OpenAI models (especially GPT-4 and 4o) returning excessively long or hallucinated UUID-like lists or arrays, and if this issue is more formally known or documented. I’ll also look for workarounds developers have successfully used, especially in tagging/filtering tasks like yours.
I’ll update you as soon as I have a clear picture.
Developers have indeed reported that when asking GPT models to output a long list of IDs (e.g. UUIDs or database keys), the model can produce incorrect or repeated entries. In structured JSON outputs, the model sometimes invents new IDs or duplicates existing ones instead of sticking to the provided list. For example, one OpenAI forum user noted that about 10% of the time GPT-4 would return a randomly generated ingredient ID that did not match any ID in the provided list, even though the ingredient name w