Not admitted · Duplicate extraction
Rules 1.1.0 · as of 14 Sep 2026
Owen Scott Muir · Said 25 Aug 2023 · owen-0687
Quote
Duplicate extraction
“If you are training a model on humans, and you take out the thing that some humans are really good at—having some ability to distinguish nonsense from not too much nonsense—you will get a very brutal regression to the mean.”
Generative AI trained on human data without a trust/intent filter will only ever produce average-of-human output and will keep disappointing; borderline, stated as a general rule.
The same statement in the same post was captured twice; the first row stands.