The ML Class and AI Class from Stanford are nearly over, and they have been a lot of fun. I have not had formal classes in either of these topics, but have seen similar ideas over the years of doing numerical programming and signal processing.
There were some technical glitches, but I learned a lot. The pace was good... unlike reading a book where I just skim through pretending to understand. The quizzes and homework encouraged active learning is more efficient than passive learning.
Many of the topics were things I had heard of, but never learned anything about. Some of the topics were completely new. I really enjoyed the ML class progression from linear regression to logistic to neural networks. NN are a simple idea, until you start thinking of how to train it. The jump from logistic regression to NN made a big difference in understanding.
In the AI class, much of the material was from Norvig's book, which I have been using as a paperweight for a while. The book is extremely dense with information, so it was nice to see explanations of even a sample of the material. (That isn't meant as a slight against the book either... just a challenge with reading it.)
Again, these classes were a lot of fun, and I learned a lot. Watching these world-class lecturers, and getting quick feedback on my understand would have been worth paying for. Getting the classes for free was amazing.