There are not many courses where the lecturer is a person who genuinely shaped the field being taught, and NYU's deep learning course is one of them, because it is co taught by Yann LeCun, a Turing Award winner and one of the central figures behind modern deep learning, alongside Alfredo Canziani, who handles the hands on practicum. That pairing defines the course and is the main reason to seek it out. LeCun's lectures bring a depth and a point of view that you simply do not get from an instructor working purely from secondary sources, and the course is willing to go into areas that a lot of practical deep learning courses avoid, including the energy based models and self supervised learning that reflect where serious research has been heading, so you come away with a stronger sense of the principles behind the techniques rather than just a bag of recipes. The other half of the value is Canziani's practicum, and it is genuinely excellent, because the accompanying PyTorch notebooks and sessions take the theory from the lectures and turn it into code you can run, read and change, which is exactly the bridge that so many theory heavy courses fail to build.
All of it, the lecture videos, the detailed written notes and the code, is published openly and free, which makes an unusually high quality course available to anyone willing to put in the work. The caveats are the honest ones for a course of this calibre. It is mathematically demanding and assumes you are already comfortable with linear algebra, calculus, probability and the basics of neural networks, so it is not the place to take your very first steps, it is a strong second course for someone who has the foundations and now wants real depth. The most shared versions of the material come from 2020 and 2021, and while the core ideas of deep learning do not go stale, a fast moving field means some specific topics and tooling details have moved on since it was recorded, so you should expect to top up the newest developments from elsewhere.
The mix of broad theory with LeCun's particular research interests can also feel a little uneven if you were hoping for a single straight line to follow, and because it is an openly published university course rather than a managed programme, there is no grading, no support and no completion track unless you are actually enrolled at NYU, so the structure and discipline are on you. My take is that this is one of the more rewarding free deep learning courses available for the right learner, the one who already understands the basics and wants to learn from someone at the very top of the field while still getting serious hands on practice. Use the LeCun lectures to deepen how you think about deep learning, work slowly and properly through Canziani's notebooks to make it real in code, accept that a few parts need supplementing with more current material, and you will get an education that is hard to match at any price, let alone for free.