Yeah, I've had a similar experience with coding in ChatGPT, both 4o and o1. They seem like they might be doing well, but then you get into trying to actually debug what they've written and find all sorts of strange errors and misunderstandings. They also both fall into weird phases of ignoring new prompts and just spitting out the same code. Hugely frustrating and a massive waste of time. I have my eye on magic.dev right now, and I'm hoping they get it right for coding, but we'll see...