From Words to Vectors: How LLMs Actually Understand and Generate Language

About this event

After our last event, the feedback was clear: people want to learn more about LLMs — and how to bring them into their own workflows. But before we can use them well, we need a real feel for how they work under the hood. So this session goes back to the roots. A machine doesn't see words — it sees numbers. We'll follow that whole journey in plain English: * **Encoding** — how raw language gets turned into numbers a model can work with (tokens and embeddings), and why that first step shapes everything after it. * **Representation** — how models learn to *understand* a sequence: from early recurrent networks, LSTMs, and GRUs that tried to "remember" context, to where they hit their limits. * **Generation** — the breakthrough that changed everything: attention and the Transformer, and how that one idea opened the door to the GPTs and LLMs we all rely on today. The goal is intuition first — enough under-the-hood detail to make the "oh, *that's* why it works" moments click, with no deep learning background required. Just curiosity. Once the foundation is down, future sessions get to the fun part: actually putting these models to work. Got a question about AI you'd like us to tackle? Drop it in the comments — we'd love to fold it into the discussion.