Read Max • 3082 implied HN points • 13 Mar 26
- Blind A/B quizzes don’t measure writing quality so much as the rough heuristics people use to guess whether text is human or AI.
- People prefer what they think is human-written, so misattribution drives apparent preferences more than any intrinsic superiority of AI text.
- The quizzes feed a stylistic arms race: readers change the “tells” of humanness and AI models keep optimizing to mimic or beat those signals.