This was predictable from the get go and many had had pointed that out already that soon people will start noticing that LLMs aren't as magical as they thought once the initial awe wanes away.

Don't think OpenAPI is doing anything here, it's not in their interest to reduce the "quality" even there's no objective and repeatable way to measure the quality either.

It's all probabilities all the way down. Who knows what the model will do. I mean, you can dry run by hand but even on quad core processors, it's damn slow so imagine the inference by hand.