Yeah. Models haven't really improved much from last year to this year.

I really love LLMs for debugging and rubber ducking, but I kinda want to write all my code.

LLMs tend to have a hard time understanding composition.