What's this about?
Mollick's note to anyone submitting an application or paper where one person reads many submissions: it is obvious in most cases, even when the language has been changed, because the ideas and concepts start to rhyme as you read through a stack.
What exactly is the detection mechanism?
Not stylistic tells in any single document — those can be edited away. It is convergence across documents. Ask a model the same kind of question a hundred times and it reaches for the same framings, the same examples, the same three-part structures. A reader seeing one submission notices nothing; a reader seeing two hundred notices they are reading the same essay repeatedly.
What's the community saying?
502 likes
- Corroboration from people who read admissions files, grant applications and journal submissions in volume.
- The link readers drew to the AI-blindness essay above: same phenomenon at different scales — sameness within one document versus across many.
- A practical implication for the "how much of HN is AI" analysis — human readers detect at the population level, which is exactly where detectors work best too.