Demystifying Large Language Models

Large language models (LLMs) have taken the world by storm over the past year or two, sparking a huge number of LLM-powered tools, products and start-ups. However, despite being readily accessible to us all, how these models work remains a mystery to many of us. This talk will demysitify LLMs by providing an intuitive understanding of how they build upon key prior advances to successfully cross the uncanny valley of text generation, and provide an overview of the surprising range of other tasks mastering this "one weird trick" seems to let these models perform. Participants will gain an intuition of how LLMs work by immersive participation in a human-powered "large audience model", as well as the opportunity to pit the latest LLMs against our very own RSECon Turing test. This talk will provide participants with a conceptual understanding of how LLMs work, the history of prior advances that made them possible, and what makes these models "foundational". It will also explore some of the strengths and limitations of these models, including examples of where they do surpisingly well and where they can struggle with suprisingly similar tasks. This talk won't make you an expert on word embeddings, deep neural networks or transformers, but it will give you a peek behind the curtain of how these seemingly magical models work and hopefully give you some tools to explain them to friends and family in an intuitive way. A recording of this session is available on YouTube: https://youtu.be/NnXCdG6nsWE