Experiments shouldn't tell you what to think, they should only tell you what not to think.

· Bits and Bobs 4/7/25
  • Experiments shouldn't tell you what to think, they should only tell you what not to think.
    • If an experiment performs better, but you can't state the mechanism of action, then be careful.
    • Maybe you're optimizing for a proxy that has decohered from the underlying thing you care about.