In an interview at NeurIPS, Jeff Dean, Google's chief scientist and co-tech lead of Gemini, discusses the new seventh-generation TPUs, highlighting their improved performance and energy efficiency, especially for low-precision floating-point formats. Dean recounts the origins of TPUs, driven by Google's internal needs for efficient machine learning computations, and emphasizes the co-design opportunities between hardware and software development. The conversation shifts to academic research funding, with Dean advocating for a vibrant ecosystem and alternative funding models like the Laude Institute. Dean expresses his passion for AI applications in healthcare, envisioning a future where every past decision informs future ones, and touches on the balance between publishing research and leveraging it in Google's products, citing the Titan model as an example of early-stage research shared with the community.
Sign in to continue reading, translating and more.
Continue
