Stanford CME 295 lectures on how LLMs are built (Stanford Online)

CME 295 is a Stanford course on the foundations of large language models. Its lectures are a good way to understand how today's LLMs are built and how they work, including a fair bit of the math that helps you build intuition from first principles. A practitioner who has taken at least one LLM foundations course every year since 2023 still found this one fresh. They say the instructors, Shervine and Afshine, explain seemingly complicated ideas simply and cover recent advancements. It suits anyone who wants to get up to speed on LLM basics, and it is especially useful if your work is mostly applied and you want to reconnect with the foundations. A tip from that learner: work through the math yourself instead of assuming what is shown is true, and use an AI assistant as a tutor when you get stuck. It makes a good mental workout. Next step: find the CME 295 lectures on Stanford Online and start with the first lecture. Revisit them from time to time as the ideas sink in.