Inside the Box: The Real Intelligence Behind Today's AI

Generative AI is often dismissed as “just a chatbot” or “just autocomplete”. The problem is that it is already far more capable than most people realise. In this pub talk I will separate hype from reality, using a couple of simple demonstrations of reasoning and generalisation in modern language models. Then we will discuss what these capabilities imply, both for what they suggest about human reasoning and language, and for what role we want AI to play in society.

Have you ever asked an AI chatbot to write you a poem, draft a tricky email or plan your weekend getaway? If so, you have probably noticed that it can do some pretty cool things. In this talk, we will pull back the curtain for a glimpse into how this technology works. Behind today’s most popular AI chatbots, you will find a Large Language Model, a mathematical engine designed to learn from and generate human language. We will explore the "box" these chatbots live in and how they handle the unpredictable real world, uncovering where the current boundaries lie and how we might move them.