How does the Genie World model work?

Genie generates interactive 3D environments from single text prompts. It reverse-engineers intuitive physics by learning from millions of videos, allowing users to control generated worlds with keyboard inputs.

What is the difference between Genie and traditional 3D rendering engines?

Unlike engines like Unity or Unreal that require programmed physics, Genie generates 2D images on the fly based on its learned understanding of the world from video data, effectively reverse-engineering physics intuitively.

How will AI models like Genie contribute to robotics?

World models like Genie help AI understand the physical world with intuitive physics, which is crucial for robotics. This understanding allows AI systems to interpret language commands and translate them into physical actions for robots.

What are the key challenges in achieving Artificial General Intelligence (AGI)?

Key challenges include replicating human-like intuitive leaps and creativity, achieving consistent PhD-level performance across all tasks, and enabling continual learning in AI systems. Demis Hassabis estimates AGI is 5-10 years away.

When can we expect AGI to be a reality?

Demis Hassabis estimates that we are about five to 10 years away from having an AGI system capable of human-like breakthroughs and consistent performance due to missing capabilities like intuitive reasoning and continual learning.

How is AI being used to accelerate scientific discovery?

AI is used in areas like material design, fusion reactor control, weather prediction, and solving complex math problems. Models like AlphaFold have revolutionized protein folding, and future AI is expected to be the ultimate tool for science.

What is Isomorphic and its role in drug discovery?

Isomorphic is a spinout company using AI to revolutionize drug discovery. Building on AlphaFold, it aims to design chemical compounds with fewer side effects and significantly reduce discovery time from years to weeks or days.

Will AI increase or decrease energy demand?

While training frontier models increases demand, AI is also driving significant efficiency gains (10-100x) in serving models. Ultimately, AI is expected to contribute more positively to energy and climate solutions than it consumes.

Inside Google DeepMind: AGI, Robotics, & World Models Explained - Demis Hassabis

All-In Podcast

Entertainment4 min read32 min video

Sep 12, 2025|574,733 views|8,575|386

chamath david sacks david friedberg jason calacanis all in podcast tech news politics big tech antitrust election covid

Save to Pod

Key Moments

TL;DR

Google DeepMind's Demis Hassabis discusses AGI, robotics, world models like Genie, and AI's role in scientific discovery.

Key Insights

Google DeepMind is the AI engine for Google and Alphabet, integrating cutting-edge models into billions of user products.

Genie 3 is a novel world model that generates interactive 3D environments from text prompts, demonstrating AI's understanding of physics.

Advancements in AI are crucial for robotics, enabling intelligent agents to understand and interact with the physical world.

AI is accelerating scientific discovery, with potential to revolutionize fields like drug discovery through projects like Isomorphic Labs.

True AGI requires creativity and intuitive leaps, not just pattern matching or incremental progress, and is estimated to be 5-10 years away.

AI tools are democratizing creativity while also empowering professionals, leading to increased productivity and new forms of entertainment.

GOOGLE DEEPMIND: THE ENGINE OF AI INNOVATION

Demis Hassabis, CEO of Google DeepMind, leads the company's AI efforts, merging various AI divisions across Google and Alphabet. He describes DeepMind as the "engine room" for these entities, responsible for developing and integrating advanced AI models like Gemini into nearly every Google product. This integration impacts billions of users daily, spanning AI overviews, the Gemini app, and Workspace applications, enabling rapid deployment of research to a massive audience.

GENIE 3: GENERATING AND UNDERSTANDING THE PHYSICAL WORLD

A significant development discussed is Genie 3, a revolutionary world model that generates interactive 3D environments from simple text prompts. Unlike traditional 3D rendering engines, Genie 3 learns intuitive physics by observing millions of videos, enabling it to create dynamic, explorable worlds on the fly. Users can interact with these seamlessly generated environments, demonstrating a profound AI understanding of physical dynamics, which is critical for applications like robotics and AR assistants.

ROBOTICS: UNLOCKING PHYSICAL INTERACTION WITH AI

The development of world models like Genie is seen as a crucial step towards advanced robotics. Hassabis highlights the potential for multimodal AI systems, like Gemini Robotics, to interpret language commands and translate them into physical actions for robots. This vision includes a potential "Android play" for robotics, akin to an operating system layer, enabling a proliferation of intelligent robots for various tasks, with debates continuing on the optimal form factors, including humanoids.

AI AS A CATALYST FOR SCIENTIFIC DISCOVERY

Hassabis's lifelong ambition is to use AI to accelerate scientific discovery, a mission spearheaded by Google DeepMind. Beyond the groundbreaking AlphaFold for protein folding, AI systems are being applied to material design, fusion reactor control, weather prediction, and complex mathematical problems. The goal is to tackle problems intractable for humans, with Isomorphic Labs aiming to revolutionize drug discovery by significantly reducing the time and cost involved through advanced AI platforms.

THE PURSUIT OF ARTIFICIAL GENERAL INTELLIGENCE (AGI)

Achieving true AGI remains a significant challenge, requiring capabilities beyond mere pattern recognition or incremental progress. Hassabis emphasizes the need for AI to exhibit genuine creativity, make intuitive leaps akin to human scientists, and possess consistent intelligence across diverse tasks. He estimates that a fully capable AGI system is likely five to ten years away, contingent on breakthroughs in reasoning and continuous learning, rather than just scaling existing models.

THE DEMOCRATIZATION AND SUPERPOWERING OF CREATIVITY

Emerging AI creative tools, such as the image generator Nano-Banana, are democratizing creativity by lowering the barrier to entry for users. These tools enable individuals to generate content simply by describing what they want, bypassing complex software. Simultaneously, these tools are empowering professional creators, like filmmakers and artists, by exponentially increasing their productivity and allowing for rapid iteration of ideas, potentially leading to new forms of co-created entertainment and a "golden age of science."

HYBRID MODELS AND SCALABILITY IN DRUG DISCOVERY

Isomorphic Labs utilizes hybrid AI models, combining probabilistic learning with known scientific rules such as those in chemistry and physics. This approach, exemplified by AlphaFold, allows AI to learn from data while incorporating essential constraints, optimizing the drug discovery process. The aim is to move from years-long discovery cycles to mere weeks or days, with early preclinical candidates expected soon through partnerships with major pharmaceutical companies.

ENERGY EFFICIENCY AND THE FUTURE OF AI COMPUTING

While AI's energy demands are a growing concern, DeepMind focuses on creating highly efficient models. Techniques like distillation enable smaller, faster models to achieve performance comparable to larger ones, leading to significant efficiency gains. Despite the ongoing trend of training larger frontier models, the efficiency improvements in serving AI are substantial. Ultimately, AI is expected to contribute more to energy and climate solutions through optimization and discovery than it consumes.

OUTLOOK: A DECADE OF TRANSFORMATION

Looking ten years ahead, Hassabis anticipates the arrival of full AGI, heralding a new renaissance. This era promises unprecedented advancements across all fields, from energy solutions to human health. The integration of AI into creative processes and scientific endeavors suggests a future where complex problems are solved more rapidly and creativity is amplified, leading to a profoundly transformed society and a new golden age of scientific exploration and innovation.

Mentioned in This Episode

●Software & Apps

●Companies

●Organizations

●People Referenced

Common Questions

Google DeepMind is the engine room of Google and Alphabet's AI efforts, merging various AI teams. It focuses on cutting-edge research and shipping AI models, like Gemini, across billions of users and products.

Topics

Genie Energy Consumption AI

Mentioned in this video

softwareGemini Robotics Models

Fine-tuned Gemini models with extra robotics data, enabling robots to interpret language instructions into motor movements.

softwareVO

A text-to-video model used by filmmakers, including Darren Aronofsky, to accelerate creative processes.

softwareKai's Power Tools

Mentioned as a historical graphics tool that has been surpassed by modern AI capabilities.

personKing Charles

Mentioned as having bestowed a knighthood upon Demis Hassabis.

personDave Matthews

Mentioned as an example for personalized music generation.

personDarren Aronofsky

Director collaborating with Google DeepMind on using AI tools like VO for filmmaking.

companyIsomorphic

A spinout company founded by Demis Hassabis focused on revolutionizing drug discovery using AI, building on AlphaFold.

companyNovartis

A pharmaceutical company partnering with Isomorphic on drug discovery.

softwareBryce 3D

Mentioned as a historical graphics tool relevant to the evolution of creative software.

softwareGenie

A new frontier for world models, capable of generating interactive environments from text prompts and reverse-engineering physics.

movieBraveheart

toolUnix

Found this useful? Build your knowledge library

Get AI-powered summaries of any YouTube video, podcast, or article in seconds. Save them to your personal pods and access them anytime.

Try Summify free