From Word2Vec to Attention: The Idea Behind Every LLM

About this event

Last time, we gave every word its own vector — "king," "queen," "dog" all became points in space. But there's a problem we quietly swept under the rug: what about a word like "bank"? In "river bank" and "bank account," the word means two completely different things — yet Word2Vec hands it the exact same vector both times. That's clearly wrong. So how does a machine tell the difference? Answering that question turns out to be the single breakthrough behind every modern LLM: attention. In this session, we'll pick up right where we left off and make the leap from static word vectors to the contextual understanding inside ChatGPT and every model like it. We'll walk through, intuitively: • The problem with static embeddings — why one fixed vector per word isn't enough • Contextualized embeddings — the big idea that a word's meaning should shift based on the words around it • Attention, in plain English — how each word "looks at" its neighbors and pulls in the meaning it needs (using the classic puzzle: in "the animal didn't cross the street because it was too tired," how does the model know what "it" refers to?) • Why this is the whole ballgame — how this one idea, "Attention Is All You Need," became the engine inside every LLM you use No heavy math — this is built to be intuitive and beginner-friendly. Come with curiosity and your questions. We'll wrap with an open Q&A (\~5 min) so you can ask whatever's been nagging you about how these models actually think. Missed the Word2Vec session? No problem — we'll do a quick recap so you're not lost. Great for beginners. Bring your questions.