Eric S. Raymond take on AI

I talked about this with Microsoft Copilot and it said that it works like a mirror of your patterns so if you are a senior dev and have good conversations about code it will have good conversations about code, if you are confused it will be confused. Like I think in words like "self-object" and "egregore" and I don't need to explain them to LLMs, they just talk like that. If you had a different vocabulary they would use that.

The times when AI models seem to be disobeying me are when I don't understand the problem and I am writing prompts that don't really make sense and instead of confronting me about it it goes along and acts confused. I think a lot of people who struggle don't know how to code, don't understand the subject matter and don't think clearly and most of all don't know what a good answer looks like.

I regularly write a paragraph about my understanding of the situation, concerns I have, say "there is some code over there which is an example of how we do this" and then end with "does this make sense? do you have any questions for me?" and then go back and forth a few times and it especially does well in handling those corner cases, like it will ask about problems that were on my mind and that I didn't get in the prompt and it will also tell me about problems I wasn't thinking of.