This session offers an accessible exploration of the inner workings of AI systems like ChatGPT, demystifying how these advanced language models generate human-like responses. We will break down the process of how large language models (LLMs) construct answers token by token, illustrating the step-by-step mechanics of language generation. Attendees will learn about the key components that drive these systems, including the role of training data, neural network architectures, and the probabilistic nature of response generation. We will also discuss how external factors, such as input prompts and fine-tuning, influence the model’s outputs, and explore the dependencies that shape their performance. Designed for a general audience, this talk avoids technical jargon while providing a clear, engaging look at the fascinating technology behind AI language systems, offering insights into their capabilities and limitations.