AI Writing Has Thousands of Tics; a Human Matching Them All Is Near-Impossible

Afinetheorem · x · 2026-11-16

The author explains the technical case for AI-text detection: models carry thousands of stylistic tics that can be extracted given millions of AI-writing samples; the odds a human exhibits that exact combined set is essentially zero. He also offers a public bet: if Bender or Gebru can produce a single 1,000+ word article by an African fiction writer, provably written before 2020, that Pangram or GPTZero flags as "100% AI," he will donate $1,000 to a charity of their choice — noting he won't lose that bet.

Original post →

More from Fun

Fun channel →