Post

“That anyone uses LLMs to code is a testament to just how bad tooling and languages are.” geohot.github.io//blog/jekyll...

I don’t agree with some of the conclusions mainly because I think that reliance on LLMs to generate code will lead to programming languages that are optimized for machine use, not human use—and humans won’t be able to easily directly use them anymore.

If there is one constant in human development… it’s passing off a task to ‘technology’ and then forgetting the skill required to complete that task without the technology.

This is why I think "AI" is not like any other technology ever made—and inherently problematic as we can't understand it (not "it's difficult" or "it's hard" but it's "it's impossible to understand").

I think id distinguish between how understanding LLMs work and understanding how LLMs arrived at their output. The former, id argue, is understandable. The latter is not.