The LLM might be terrible at writing machine code directly. The kinds of mistakes I see GPT-4 making in Python, PostScript, or JS would be a much bigger proble

The LLM might be terrible at writing machine code directly. The kinds of mistakes I see GPT-4 making in Python, PostScript, or JS would be a much bigger problem in machine code. It "gets confused" and "makes mistakes" in ways very similar to humans. I haven't had a chance to try DeepSeek R1 yet.