It feels kind of crazy to me that AlphaFold works.

· Bits and Bobs 11/4/25
  • It feels kind of crazy to me that AlphaFold works.
    • But maybe the reason AlphaFold works isn't that unrelated to why transformers are good at images.
    • The easiest way to predict which way an image is oriented is by developing a world model that picks up on subtle cues that humans would have a hard time even describing.
    • The easiest way to predict which way the protein will fold is by developing a world model that picks up on subtle clues that humans would have a hard time even describing.
    • It's hard for our brains to handle more than 2 dimensions.
      • But in tensor space it doesn't matte.
      • They can handle arbitrary dimensions that we can only do in 2 or 3.
    • Apparently DeepMind decided to tackle protein synthesis when they heard there was a game to predict folding of proteins that humans could play.
      • That implied there was some subtle correlation, that transformers could exploit even more directly.
    • For pattern recognition, if humans can do it transformers can do it.