Skip to content

Article

Bitter Lesson Engineering

danielmiessler.com

Source ↗ ← All highlights
  • We have to learn the bitter lesson that building in how we think we think does not work in the long run.
  • It’s so seductive to think that we humans have the magic that AI will never have, but that magic is often just hubris.
  • Another big part of this is making sure your AI has the best tools available. If you’re trying to get help managing your life, like managing routines, improving your finances, etc: don’t give it a bunch of your own accumulated methodologies and tell it to implement them. Instead, articulate exactly the life you want to have.
  • If you’re getting help starting a business, don’t tell it how to help you: tell it the business you want to build and the life you want to have. Focus less on the steps of execution and focus more on the results you want and don’t want.
  • The smarter the model, the less “how” scaffolding you need.
    1. The way we think about logic and intelligence and efficiency are very likely primitive
    2. So we shouldn’t be hard-coding those rules or ideas into how we “teach” AI to do things
    3. As AI gets smarter it’ll come up with way better ways to do the same thing from first-principles
  • Don’t confuse the “what” with the “how”. Be extremely specific about what you want, and then give the best tools you have to the best AI you have, and let it figure out how to execute.
  • This means as the AI gets smarter, our scaffolding becomes more about preferences than execution, ultimately making our entire system meta-upgradeable instead of BLE-hobbled.
  • A BLE-hobbled system is one where the scaffolding has aged to the point of making your overall system worse instead of better. After which point the AI could actually do a better job if it didn’t have to follow our super-smart, dumb instructions.