This is Jeff Dean’s Ken Kennedy Institute Distinguished Lecture, delivered at Rice University on 13 February 2024 and posted on the official Rice Ken Kennedy Institute YouTube channel. Dean joined Google in 1999 and serves as its chief scientist, focusing on AI advances across Google DeepMind and Google Research, and the talk reflects his vantage point on the systems side of machine learning.
In his own framing, Dean argues that a combination of improved algorithms and major efficiency gains in ML-specialized hardware now lets researchers build far more capable, general-purpose systems than before. He uses the Gemini family of multimodal models as a running example of what these advances make possible, and then turns to applications, highlighting machine learning in science, engineering, and health.
This is a practitioner-level survey that rewards viewers who want the hardware-and-systems perspective rather than only the modeling story. As a primary source it is a firsthand account from one of the people most responsible for Google’s machine-learning infrastructure, including the TPU effort.
There is no jeff-dean person entry in the library yet, so this talk links to the organizations and concepts closest to its content.