Gemini and the path to AGI with Koray Kavukcuoglu

Gemini and the path to AGI with Koray Kavukcuoglu

More

Summary

In this episode of the Google DeepMind podcast, host Professor Hannah Fry speaks with Koray Kavukcuoglu, who now leads the development of Gemini from California after years in DeepMind’s London offices. They reflect on how the organization has changed since the days of separate teams for AlphaGo, AlphaStar, and AlphaFold, and why a single general model has replaced many narrow projects.

Kavukcuoglu explains how Google DeepMind decides what to pursue by balancing three forces: AI research ambitions, products that reveal what people need, and science as a driver of progress. He discusses why he sees products as a route to AGI, and how the same approach can be applied to scientific domains such as fusion by training models to act in those fields rather than solving one problem at a time.

The conversation also covers the limits of current models’ understanding of the physical world, including fluids, airflow, heat transfer, and collisions, and his expectation that models will become numerically accurate enough for engineering work. They also look at the energy cost of training frontier models compared with biological intelligence, and why the goal is helpful digital intelligence rather than a replica of the human brain.


📺 Source: Google DeepMind · Published November 19, 2026
🏷️ Format: Podcast

1 Item

Channels

1 Item

Companies