A Decade Inside Language Models: From Word Embedding to Today's LLMs

Presented by Joseph T. Yun, this talk draws on Yun's research journey—beginning when Google first published word embeddings during his graduate studies—to give an accessible look at how modern LLMs work. We'll build intuition around core ideas like training data, token prediction, and the limits of what generative models can do, with the goal of helping you sense where these tools tend to excel and where they fall short.