AI Safety (CS 2881) Fall 2026 Lecture 3: Modern LLM Training and Inference

This lecture from CS 2881, the graduate AI Safety course taught by Boaz Barak at Harvard in Fall 2026, explains how modern large language models are trained and how inference works. The course covers challenges in the alignment and safety of AI, considering both technical questions and the societal impact of the field, and its lectures are recorded and published online. A clear, current university-level explanation of the LLM training and inference stack for developers who want the fundamentals behind the tools they use.