Not Your Typical AI Talk

Advanced architectural design has long had an intimate relationship to developing technologies and to contemporary mathematics. This has been true historically from the earliest developments in the visual depiction of space, through Gaspar Monge’s invention of descriptive geometry, and into the earlier 21st century with advances in the description and manipulation of complex curvature. While long in duration, the nature of this relationship has been unclear following recent advances in widely-available generative AI. The content of this lecture will focus on some of the fundamental principles at work in a functioning model. Why is vector mathematics particularly important to AI? How do language models understand the semantic content of words and images? What is meant when we say models are “reasoning” or “making decisions?”

Jean (you can call him John) is a research scientist at Google DeepMind in New York on the Gemini Large Scale Pretraining team. Previously, he led a small research team within the Algorithms & Optimization group within Google Research. Before joining Google, he completed a PhD in Computer Science at Harvard University, advised by Edoardo Airoldi and Salil Vadhan, where he was a 2017–2018 Siebel scholar. Prior to that, he completed his undergraduate degree at École Polytechnique in Paris. At the time, he looked into using neural networks to generate new distributions from data.