Some good piece of high-level code is an accurate impression of the logic and theory for moving some bytes around, so it is most human-engineer-parsable by design. Parsing it with GPT creates even higher-level abstraction, much like trying to explain some code to a kid – I think this can be somewhat good for an elementary school application - but it's not serious to use for a developer or a business owner - who would rather be equipped with N language syntax dictionary and business context to parse and understand a piece of code than rely on a probabilistic machine to chew it for him.