In my experience 4o is really good at ignoring user-provided corrections and insanely regurgitating the same code (and/or the same problems) over and over again.

ChatGPT 4 does much better with corrections, as does Claude. 4o is a pox.