I have a theory about creation with AI.

(wait! please come back)

My gut says this can’t be original, but I haven’t seen it exactly expressed anywhere.

  1. The best art (and I’ll extend this to products in general) reflects the taste and discernment of just one person, or at least very few people.
  2. LLM’s represent the antithesis of that, the voice of the masses, and perhaps worse: the voice of the masses digested by a deflavorizing machine.
  3. The skill in creating with LLM’s is then to remain in charge, to make sure YOUR taste and discernment are dominant. If this effort gets in the way of creating, then the LLM is harmful to your endeavor.
  4. Code has the opposite dynamic! You generally want the most generic, boring, un-creative implementation of a thing, unless you’re trying to break new ground in computer science. The voice of the masses is an asset.

I think one way to navigate #3 (at least for digital projects) is tool building, using AI not to necessarily make the thing, but make to make peripheral tools that improve your life or make creating more productive or pleasant. For SpinDoc, I don’t plan on using AI for graphics, game design, level design, etc– but the level editor is a place where “taste and discernment” don’t matter so much (at least not yet), and letting the LLM do most of the work on that has been helpful to me.