The Generative AI Laboratory led by Ethan Mollick at the Wharton School of the University of Pennsylvania has released a tool for teachers called Primer.
Instead of directly generating a bunch of exercises for you, it first "interviews" the teacher: what is being taught in this class, what level the students are, what they already know, and where they are most likely to make mistakes. You can speak or type, and lesson plans and scoring rubrics can be uploaded directly.
It then writes a design document and does nothing until the teacher approves it. The demonstration uses a lesson on the water cycle for seventh and eighth grade students.
After students enter, they will encounter three AI characters: a tutor to find out how much they know first; a practice partner; and a coach to give feedback. There are two ways to play the exercise, either role-playing with a character who has his own goals, or explaining knowledge points to a character who "hasn't understood yet".
The design of "Tell others to listen" is quite clever. Being able to explain a concept clearly is a better indicator of true understanding than answering ten multiple-choice questions correctly.
The teacher can get reports for each student and the whole class. The prompt words, code and papers are all open source and can be run on local models. Schools do not have to worry about student data being transmitted outside.
Wharton himself made it clear: This is an experimental prototype and has not been tested in a real classroom. How effective it will be, we won’t know until a teacher actually uses it in class.