LLM University is one of those free resources that quietly turned out to be better than a lot of paid material, and the reason is that Cohere clearly put teachers rather than marketers in charge of it. The thing it does best is the retrieval and representation side of large language models, which is exactly the part most beginner courses gloss over. It takes the time to explain what an embedding really is, how semantic search works once you have those embeddings, and why retrieval augmented generation is the practical answer to a model that does not know your private data. Those explanations are written plainly, with diagrams that earn their place, and they stuck with me more than some far longer treatments.
The modules are short and self contained, so you can treat the whole thing as a reference and drop into the bit you need, whether that is prompt engineering, fine tuning, or deploying something to production. The build-alongs are where it pays off, you actually put together a RAG based chatbot and a multi-turn agent rather than just reading about them. The obvious caveat is that the examples run on Cohere's own API, which makes sense given who built it, so if you work with another provider you will be translating calls as you go, though the underlying ideas are provider agnostic and that translation is rarely hard. The other small frustration is that the curriculum has been reorganised more than once as the field moved, so depending on when you arrive the recommended path can feel slightly stitched together rather than one clean line from start to finish.
And while it is approachable, it assumes a developer audience, so a complete beginner with no coding background would be happier starting somewhere gentler. Set against the fact that it costs nothing and has no ads, those are easy trade offs. If you want to genuinely understand embeddings, search, and RAG instead of just memorising prompt tricks, this is one of the best free places to do it.