What's this about?
Mollick acknowledges that instructors never asked for any of this, then says plainly that if you teach and have not updated your expectations in the last six months, three things have changed underneath you: what AI can do (a lot more), its error rate (a lot lower), and its ubiquity among your students (near total).
What exactly is the "stale picture" problem?
Most people's mental model of these tools is frozen at whichever version they last tested, and assessment design is built on that frozen model. An assignment that was AI-proof a year ago because models could not do it may now be trivially solvable — and the instructor has no signal that anything changed, because the assignment still looks the same.
What's the community saying?
653 likes
- Sympathy for the position teachers are in, which Mollick leads with and which many replies echo.
- Argument about whether the response is redesigning assessment or changing what is taught, with no consensus.
- It connects to his separate observation from yesterday's digest that markers reading many submissions *can* tell — detection works at the stack level even when it fails per-document.