Some thoughts on teaching AI

Extraction of intelligence is the key thing: how to develop the instincts, infrastructure, and persistence to get really great things out of the model.
Understanding how computers work, and having some familiarity with Git, makes it much easier to set up that infrastructure. It is less about comprehensive know-how, because you can ask the model how to do something, and more about having enough foundation to think creatively about what is possible.
Teaching people what is possible is easier than teaching people how to do specific things.
The extraction part is really fun.
This is an incredible gift of intelligence and time. What will students, and professors, do with it? Can lectures become a lot more exciting? Can classes become more hands-on and more innovative?
If a student uses a chatbot to one-shot an essay, or Claude Code or Codex to one-shot a problem set, that is tough, and most will do this. But if a student iterates and iterates until the final product is pristine, that is learning.
The social elements of academia should be invested in even more. We are the primary consumers of our work and ideas, so we need more forums to share the new awesome things the models are helping us create.
The bar has to get a lot higher. Who can make something great with the model? This clashes with Q-score incentives and grade inflation.
Honors theses need to become amazing.
Right now, someone with a lot of experience using the top models to build things can get dropped into almost any company and become a huge value add. This will not always be the case as labor skills converge and as models become more autonomous.
Force students to write in class. Use handwritten exams. Require oral presentations. Make them say out loud what the model does and how it does it. Have them explain it to you and to their peers, even if the model taught them what to say. Out-of-class assignments are for the experience, not the assessment.
Find hobbies and creative outputs where you write.
But a lot of this will end up being on them. The ones who want to learn will learn a lot. Absolute Superman knows what's up.¹

Superman page about choosing your own path

Pretty clear at this point that we do not have the answers yet.

This will not scale. ↩