Deep Dive into Long Context

Google DeepMind SVP and Chief AI Architect Koray Kavukcuoglu joins host Logan Kilpatrick to reflect on the journey from DeepMind's early reinforcement learning milestones to Gemini and what it means to build AGI at Google.

Carolina Parada, Stuart Bowers, Kanishka Rao, and Jie Tan from Google DeepMind join host Logan Kilpatrick inside the Gemini Robotics Lab to introduce Gemini Robotics 2, Google DeepMind's new suite of models bringing whole body intelligence, dexterity, and multi-robot collaboration to general purpose robotics. Their conversation covers the arc from reinforcement learning to vision language action models and the open challenge of physical AGI.

Sameer Samat, President of Android at Google, joins host Logan Kilpatrick to discuss Android 17 and what it means to shift from operating systems to intelligence systems. Their conversation covers why Android didn't say the word AI once at the Android Show, the new Gemini Intelligence feature suite including Rambler, app automation, Superfill, and generative widgets, and what it looks like when Gemini can see the road with you in a live demo inside a Volvo EX60.

Yossi Matias, Vice President at Google and Head of Google Research, joins host Logan Kilpatrick for a wide-ranging conversation about what it means to do research at Google scale. Their conversation explores the "magic cycle” of research from breakthrough to real-world impact across flood forecasting, MedGemma, and the just-launched Gemini for Science. Learn more about AI tools for scientific discovery including Co-Scientist and Empirical Research Assistance (ERA), and topics like speculative decoding, generative UI, and quantum computing milestones including the Willow chip.

To mark the launch of Gemini 3.5 Flash, Logan Kilpatrick sat down at Gradient Canopy with four of the people who built it: Jeff Dean, Koray Kavukcuoglu, Noam Shazeer, and Oriol Vinyals. They talked about the origin of the Gemini project, the bet on a single unified model, why each Flash generation now outperforms the previous Pro, and what's coming next.