According to Jeff Dean (@JeffDean), both Gemini Flash and Waymo are leveraging knowledge distillation, as detailed in the research paper arxiv.org/abs/1503.02531, to create high-quality, computationally efficient AI models from larger-scale, more resource-intensive models. This process allows companies to deploy advanced machine learning models with reduced computational requirements, making it feasible to run these models on resource-constrained hardware such as autonomous vehicles. For businesses, this trend highlights a growing opportunity to optimize AI deployment costs and expand the use cases for edge AI, particularly in industries like automotive and mobile devices (source: twitter.com/JeffDean/status/1998453396001657217).