akshith

Learning when AI can already do it

Published

I’m trying to figure out what it means to learn deeply when AI can already do the work. Without getting too meta about what it means to learn or work, I want to figure out how I want to use AI for myself.

I don’t recall ever saying, “I want to figure out how I want to use the internet or cryptocurrency for myself,” but raising this question about AI doesn’t feel off yet. Should it? The answer might just be, “This isn’t different; use it and misuse it responsibly.” But if that’s not the case, why not?

I don’t have an argument yet. I’m still reading what others have to say.

Reading

  1. Terence Tao on AI in mathematics (and beyond) frames the problem around selective friction, verification, and a deliberate “AI diet.”
  2. Should you use AI for a task? by Bruce Schneier distinguishes work, where the output matters, from the cognitive gym, where doing the work is the point.
  3. How To Write With An LLM by Thomas Ptacek applies that distinction to writing: write the draft, then use the model as a critic rather than a ghostwriter.