MIT runs this as a one-week intensive in January, and then does something I wish more institutions did, it puts the entire thing online for free, lectures and labs and all, and refreshes it every year so the content keeps pace with a field that changes fast. The lectures, mostly delivered by Alexander and Ava Amini, take you from what a neural network and gradient descent are through convolutional networks for vision, recurrent and attention-based models for sequences, and on into generative models and the deep learning behind the tools everyone is now using, and the production quality is high enough that it never feels like a grainy recorded lecture. What lifts it above a YouTube playlist is the software labs, which run in Google Colab so you do not have to set anything up, and which have you implementing the ideas rather than just watching them, building models and seeing them train. The honest caveat is that this is genuinely MIT-paced, it assumes you can already write Python comfortably and that a bit of calculus and linear algebra will not scare you off, and because it was built as a compressed intensive it moves through serious topics briskly, so a complete beginner will likely feel the current.
That is not a flaw so much as a positioning thing, it is an introduction for people who already have some technical footing, not a gentle first step for someone who has never coded. If that is you, do the labs properly rather than just watching the lectures, because that is where the learning actually happens, and pair it with a slower explainer like StatQuest or 3Blue1Brown whenever a concept needs more time to settle. For free, current, rigorous deep learning teaching from a top institution, it is hard to do better.