A free online textbook tries to teach large language models without the hype.
The book, now in its third revision, organizes LLM theory into six chapters: pre-training, generative models, prompting, alignment, inference, and reasoning. Its authors explicitly say it skips a survey of the latest flashy models and benchmarks in favor of the ideas that underpin all of them. The intended audience is college students, NLP professionals, and practitioners looking for a reference, not a leaderboard. Being a revision rather than a first release, it reads less like a hot take and more like an attempt to consolidate what's already been learned.
Most LLM coverage chases whichever model topped a benchmark this week, which makes it hard to build a durable mental model of how these systems actually work. A textbook that separates the stable mechanics, like attention, alignment training, and inference tricks, from the churn of model releases stays useful long after this week's leaderboard is forgotten. That is a rarer and arguably more valuable resource than another product announcement.
No press release has ever survived contact with a syllabus, and that is exactly the point.