Exciting! A spicy take on LLMs from the fiery maintainer of the Linux kernel?!
Oh, it’s just a reasonable, normal kind of take? “LLMs are a tool” is right. You can use them well or not, just like Google searching, StackOverflow, autocomplete, or high-level languages (remember the C++ vs Java debates of the 90s? No? Anyone? Even I’m not old enough to remember the “actually C is a high level language and everyone should use assembly” arguments though).
There’s arguments to be made about the ethics of the major “frontier” AI companies, certainly. And LLMs are being pushed into weird places (bad summaries of news headlines?), to be sure. But LLMs are useful for coding. We’re still figuring out how to use them right – just like we had to figure out how to balance the existence of StackOverflow and the Ctrl+C and Ctrl+V keyboard shortcuts.
To that point, I think we can have (and are having) interesting conversations around “how do we best use LLMs for coding?” That’s what I want to see more of. What are the right harnesses, skills, documentation practices, test patterns, etc. that produce better code (fewer bugs, better functionality, better user experience, better products)? That’s what I’m after, anyway.
The email is here.