The code generated by large language models (LLMs) has improved some over time — with more modern LLMs producing code that has a greater chance of compiling — but at the same time, it's stagnating in ...