This is so real. I’m forced to use it by my employer and it just sits there doing the cringiest shit all the time. Like everything is boilerplate and it’s just a deeply unsatisfying spaghetti mess of function calls. It’ll work if you have a good feedback loop, eventually, but only ever as a crude prototype.

The LLMs are just somewhere between an averaging and a lossy compression of everything on GitHub. There’s nothing about the current paradigm of “AI” that is going to somehow do better than just rehashing that training set but with the inclusion of various classes of errors.

I think it’s better to view it as spicy search rather than any form of intelligence.