Caine Prize investigated its AI-controversy winner — and exposed detector blind spots

paulnovosad · x · 2026-11-12

The Caine Prize published a transparent account of how it probed claims that this year's winning story was AI-generated. Author paulnovasod stresses it relied on "multiple lines of evidence, not a single score." The accompanying analysis shows a detector defined as "differs from X" cannot see a target that IS X — three faithful masks answered their own predicates yet none found the lens itself, illustrating how models print confident answers to questions their vocabulary can't ask.

Related event: Caine Prize Winner Flagged 99% AI-Generated, but Probe Finds No Evidence(2 posts)→

Original post →

More from Fun

Fun channel →