Archive
Notes and Cheat Sheet
Review
I was initially hesitant on taking this class since I was already enrolled in 4 techs, but I yolo’d it since Sergey only teaches once in a blue moon. He’s clearly been busy running PI in the age of foundation models.
It was a good thing that I enrolled, because RL is probably the best “intuition” class for ML. I finally understood many terms that repeatedly came up in my readings like on/off policy, GRPO, behavioral cloning. Robotics seemed to be the dominant motivation of concepts in the course, and there’s no better lecturer than Sergey when it comes to robots.
Regarding difficulty, the class requires a firm grasp of probability; the slides are full of deeply nested conditional expectations. Sergey’s sentences are also denser than most lecturers. Overall, I give it a 4.5/5 in difficulty (especially the midterm!) and 5/5 in fun. I took the class P/NP, but in hindsight I definitely should’ve taken it for a letter grade since I did great on the final paper.