Like the article says... I feel it's counter-productive to picture an LLM as "learning" or "thinking". It's just a text generator. If it's producing code that c

Like the article says... I feel it's counter-productive to picture an LLM as "learning" or "thinking". It's just a text generator. If it's producing code that calls non-existent APIs for instance, it's kind of a waste of time to try to explain to the LLM that so-and-so doesn't exist. Better just try again and dump an OpenAPI doc or some sample code into it to influence the text generator towards correct output.