This hands-on technical track covers controlling and evaluating AI systems that may intentionally subvert oversight, working through the field's primary literature in full with exercises woven between the sections.
No enrollment. Sign in only if you want progress saved.
Module 01 · 1 lesson · 6 readings
Module 02 · 3 lessons · 2 readings
Module 03 · 4 lessons · 1 reading
Module 04 · 4 lessons · 10 readings
Module 05 · 3 lessons · 5 readings
Module 06 · 3 lessons · 5 readings
Module 07 · 1 lesson
If you are async: Everything is open and there is no enrollment. The readings reproduce the field's papers and posts in full, with exercises placed where the argument turns, and the modules build on each other in order. Sign in to save your progress: every tap-to-reveal question you rate then joins a spaced-repetition deck on the Review tab, and written responses can get automated feedback on how clearly they show their reasoning.
If you are sync: Instructors can open a classroom, have learners join it, assign lessons and readings from any module, and see who has completed them. A classroom can also carry its own grading key, so every member's written responses get feedback without each person bringing their own.
We assume that you have the knowledge equivalent to an intro-level AI safety course (Bluedot, University Intro Courses)
We suggest that you have some experience reading ML or AI Safety research papers & blogposts
