On LLMs:

  1. LLMs know about literally everything that has been published in the world ever. However, they know nothing about the specifics of my work and what I want to do.
  2. I know a lot about the specifics of my work1, but I know only a tiny fraction of the total knowledge of humanity

In other words, LLMs have a lot of general context, while I have a lot of specific context.

In order for great code to be written, it’s important to have both a lot of general and specific context.

Context is that which is scarce. 2

Thus, I have two options:

  1. I can provide the LLM with specific context and have it write code for me.
  2. I can ask the LLM to give me relevant general context, and I can write the code myself.

I like option 2 a lot more. With option 1, I am just outsourcing my work. With option 2, the LLM is a resource that helps me learn and grow.

When you use AI tools, ask yourself: are you serving AI, or is AI serving you?

  1. Okay, maybe I don’t always know a lot about the specifics of my work, but I know at least a little, which is more than the LLM does, which is zero. ↩

  2. https://marginalrevolution.com/marginalrevolution/2022/02/context-is-that-which-is-scarce-2.html ↩