Jay Alammar occupies an unusual and valuable spot in AI education, because he is less a course creator and more the person who drew the diagrams that an entire generation of practitioners learned attention from. The Illustrated Transformer, The Illustrated Word2Vec, The Illustrated Stable Diffusion and the rest of the series on his blog took papers that most people found impenetrable and rebuilt them as a sequence of pictures, walking one token or one vector through the model and showing what happens to it at each step. The effect is hard to overstate. When someone new to the field asks how attention works, the honest answer for years has been go read Alammar, and that is still largely true.
What makes it work is discipline about intuition. He resists the temptation to show off the maths and instead keeps asking what is this actually doing, so you come away with a mental animation of queries matching against keys rather than a memorised formula, and that picture is what makes the real papers readable afterwards. Beyond the free blog, he has widened his reach in two directions. First, he co authored Hands-On Large Language Models with Maarten Grootendorst for O'Reilly, which takes the same visual sensibility and pairs it with real code for tokenisation, embeddings, semantic search, prompt engineering and fine tuning, and it is one of the more approachable serious books on the topic.
Second, through his work at Cohere he has been involved in free short courses on DeepLearning.AI, including the semantic search and embeddings material, which are a nice practical complement to the reading. The honest limitations are the flip side of the strengths. This is not a packaged curriculum with a start and a finish, it is a brilliant collection of explainers you have to string together yourself, and because it leans so hard on understanding it will not on its own teach you the messy engineering of running these systems in production. A couple of the landmark posts are also old enough now that the frontier has moved past some of the specifics, though the core ideas they teach have aged remarkably well.
My take is simple. If you want to actually understand what is happening inside a language model rather than just call an API, start with Alammar, and do it early, because almost every other resource gets easier once his pictures are in your head. Then reach for the book when you want to build, and layer a more production focused course on top when you are ready to deploy.