Demis Hassabis

AlphaGo: A Decade of AI Evolution Since the Go Challenge

The 2016 AlphaGo victory over Go world champion Lee Sedol marked a pivotal moment in AI, demonstrating machines could surpass human intuition and calculation in complex tasks. This triumph, fueled by reinforcement learning and deep neural networks, laid the groundwork for significant advancements in AI, including large language models and scientific grand challenges like protein folding, by proving AI's capacity to generate novel, non-human-like solutions.

Introducing Lyria 3 Pro

Insufficient Data for Knowledge Extraction

The provided source content consists of a brief social media interaction expressing gratitude. It contains no technical insights, factual claims, or substantive information suitable for structured knowledge extraction.

Gemma 4: Outperforming Larger Models

Gemma 4 demonstrates superior performance compared to models ten times its size, indicating significant advancements in efficiency and capability within smaller model architectures. This performance is particularly notable given the logarithmic scale of comparison.

Gemma 4: Powerful Performance in a Compact Model

Gemma 4 demonstrates exceptional capabilities despite its small size. This indicates advancements in model architecture and training that enable high performance within resource-constrained environments. Developers should follow official channels for updates on this efficient model.

Gemma 4: Next-Generation Open Models Launched with Diverse Sizes and Licensing

Gemma 4 introduces a new suite of open models, featuring optimized architectures for varying computational demands. These models are designed for adaptability and broad deployment, offering solutions from high-performance cloud applications to efficient edge device integrations. The strategic release under an Apache 2.0 license facilitates widespread adoption and custom development.

Gemma 4: Google DeepMind's Latest Open Models Offer Diverse AI Solutions

Gemma 4, developed by Google DeepMind, introduces a new suite of open models, including 31B dense for raw performance, 26B MoE for low-latency applications, and efficient 2B/4B models for edge devices. These models are designed for fine-tuning to specific tasks and are available under the Apache 2.0 license, facilitating broad adoption and development.

Demis Hassabis Shares Enthusiastic But Unspecified Update

Demis Hassabis, CEO of Google DeepMind, posted a message consisting solely of five diamond emojis. This cryptic post offers no explicit information but may signal significant, undisclosed progress or an upcoming announcement related to DeepMind.

Demis Hassabis Unveils Isomorphic Labs: AI for Drug Discovery

Demis Hassabis announced the launch of Isomorphic Labs, a new Alphabet company. The company aims to apply AI to accelerate drug discovery, building on the success and techniques developed at DeepMind. This initiative signifies a strategic expansion of AI research into practical, high-impact scientific fields.

Gemini 3.1 Flash Live: A Step Towards Next-Generation Voice AI

Gemini 3.1 Flash Live is presented as Google's most advanced audio and voice model to date, designed to enhance voice-first agent capabilities. It features improved latency, precision, and natural interaction. This model is accessible via the Gemini App's Gemini Live feature and through Google AI Studio for developers.

Gemini streamlined data import enhances user migration from competing AI platforms

Google Gemini is rolling out new desktop features that simplify user migration from other AI applications. These features allow users to import preferences and chat histories, enabling a seamless transition and continuity of user experience. This strategy aims to reduce friction for users switching to Gemini, addressing a key barrier in AI platform adoption.

Demis Hassabis Announces Gemini 1.5 Flash Availability

Demis Hassabis announced the immediate availability of Gemini 1.5 Flash, a new AI model designed for enhanced efficiency and performance. This release targets developers and enterprises seeking advanced AI capabilities. Further details and benchmark information are accessible via the official Google AI blog.

Gemini 3.1 Flash Live: A Step Towards Voice-First AI Agents

Gemini 3.1 Flash Live is Google DeepMind's latest audio and voice model, enhancing natural language interactions with lower latency and improved precision. This development is crucial for advancing voice-first AI agents, as highlighted by its integration into the GeminiApp and availability in Google AI Studio for developers. The model significantly improves function calling, contributing to more useful and informed AI applications.

DeepMind and Isomorphic Labs Focus on AI for Scientific Discovery

Demis Hassabis highlights that DeepMind, through projects like AlphaFold, and Isomorphic Labs are actively engaged in applying AI to scientific research and discovery. This indicates a strategic direction towards leveraging advanced AI for complex scientific problems, particularly in areas like protein folding. The collaboration suggests a concerted effort to translate AI breakthroughs into tangible scientific advancements.

Lyria 3 Pro: Advanced AI Music Generation for Enhanced Composition

Lyria 3 Pro is a new AI-powered music generation tool, now integrated into the Gemini App for subscribers and accessible via API for developers. It enables the creation of high-fidelity music compositions up to three minutes in length, allowing for detailed structuring of musical segments like intros, verses, choruses, and bridges.

AI to discover fundamental scientific theories

Demis Hassabis postulates that AI tools could significantly contribute to uncovering fundamental scientific theories. This process would involve extensive pattern processing and matching to arrive at elegant and compact explanations for complex phenomena, potentially leading to breakthroughs comparable to those of Newton or Einstein.

Starlink Verified as Useful by Demis Hassabis

Demis Hassabis, CEO of Google DeepMind, has publicly endorsed Starlink, stating it is "amazingly useful." This statement, given Hassabis's standing in the technology sector, implicitly validates Starlink's practical utility and effectiveness from an expert user perspective. No further context or specific use-cases were provided, limiting deeper analysis of its applications.

AI to Advance Fundamental Physics Discoveries

Demis Hassabis and Elon Musk discuss the potential of AI to drive future scientific breakthroughs. Hassabis posits that AI can uncover elegant, compact descriptions of the universe's deepest mysteries through extensive pattern processing. Musk suggests that future intelligence output will almost entirely focus on new creation rather than discovering basic rules, as fundamental physics is nearly complete.

DeepMind Appoints Jas Sekhon as Chief Strategy Officer for AGI Development

Demis Hassabis announced Jas Sekhon

Stitch by Google: AI-Powered Vibe Design for Rapid Prototyping

Stitch by Google is an AI-native design platform that allows users to generate high-fidelity designs from natural language descriptions. It enables rapid iteration by stitching screens into interactive prototypes and managing a portable design system. The platform also supports hands-free voice interactions for real-time layout updates and design variation exploration.

AlphaFold and AI Enable Rapid, Personalized Cancer Treatment in a Canine Model

A single individual leveraged AI tools like AlphaFold and ChatGPT to develop and administer a personalized mRNA cancer vaccine for a rescue dog. This case demonstrates the potential for rapid, democratized drug discovery and personalized medicine, significantly accelerating traditional pharmaceutical pipelines.

AlphaEvolve Advances Ramsey Number Bounds by Automating Search Procedure Discovery

AlphaEvolve, an AI developed by Google DeepMind, has achieved new lower bounds for five classical Ramsey numbers. This significant advancement in computational mathematics is attributed to AlphaEvolve's ability to autonomously discover search procedures, a task traditionally requiring human-designed algorithms. The breakthrough updates results that have remained stagnant for over a decade, demonstrating a novel application of AI in solving complex combinatorial problems.

Google DeepMind Expands London Presence with New AI-Focused Facility and Public Engagement Space

Google DeepMind is expanding its London operations with 'Platform 37,' a new building designed to foster AI breakthroughs. The facility includes 'The AI Exchange,' a public-facing space dedicated to AI education through exhibitions and events. This strategic investment reinforces DeepMind's commitment to London's talent pool and aims to enhance public understanding and engagement with artificial intelligence.

Google DeepMind Expands London Presence with New AI-Focused Hub

Google DeepMind has established a new London headquarters, "Platform 37," signaling a significant investment in the region's AI ecosystem. This facility includes "The AI Exchange," a public engagement space, underscoring a commitment to both advanced AI research and public education. The expansion reinforces London's position as a key hub for AI talent and innovation.

Demis Hassabis Discusses AlphaGo and AGI on DeepMind Podcast

Demis Hassabis, CEO of Google DeepMind, recently featured on the Google DeepMind Podcast alongside Michael Fry to discuss the Alpha series (including AlphaGo) and Artificial General Intelligence (AGI). The discussion likely covered advancements in AI for science and the broader implications of these technologies.

AlphaGo: A Decade of AI Advancement and Its AGI Implications

Ten years post-AlphaGo's victory, the AI community reflects on its pivotal role in initiating the modern AI era. The technological advancements demonstrated, particularly by "Move 37," proved AI's readiness for complex scientific problem-solving. These methods are now considered foundational for the development of Artificial General Intelligence (AGI).

Aletheia: Advancing AI in Mathematical Research from Olympiad to PhD-level

Aletheia, an advanced math research agent powered by Gemini Deep Think, demonstrates robust capabilities in mathematical problem-solving. It excels at iteratively generating, verifying, and revising solutions in natural language, extending beyond Olympiad-level problems to PhD-level exercises. The system leverages intensive tool use to navigate complex mathematical research and has achieved milestones such as autonomously generating research papers and solving open problems.

Demis Hassabis on the Path to AGI and the Impact of AI on Society

Demis Hassabis, CEO of Google DeepMind, discusses the current state and future of AI, emphasizing the need for breakthroughs in continual learning, memory, long-term reasoning, and planning to achieve Artificial General Intelligence (AGI). He defines AGI as a system possessing all human cognitive capabilities, including creativity and physical intelligence, and estimates it to be 5-10 years away. Hassabis also highlights the potential of AI in various product applications like smart glasses and addresses the economic and societal implications of widespread AI adoption, stressing adaptation and the evolving nature of human purpose.

Hassabis on an AI Shift Bigger Than Industrial Age | Bloomberg Live

Google DeepMind Bolsters Robotics Team with Boston Dynamics Veteran

Aaron Saunders, former CTO of Boston Dynamics, has joined Google DeepMind as VP of hardware engineering. This strategic hire significantly strengthens DeepMind's robotics team, signaling an increased focus on the intersection of robotics and AI. The company is actively recruiting to further expand its capabilities in this domain.

Google DeepMind Partners with Boston Dynamics, Expands Robotics Team for AGI Development

Google DeepMind is advancing its Gemini Robotics initiative to integrate AI into physical systems, a crucial step for achieving Artificial General Intelligence (AGI). This effort includes a strategic partnership with Boston Dynamics, leveraging DeepMind's robotics models with Boston Dynamics' Atlas humanoid hardware. Concurrently, Google DeepMind is expanding its internal robotics team, notably by hiring former Boston Dynamics CTO Aaron Saunders, to strengthen its hardware engineering capabilities.

Demis Hassabis on the Future of AI: AGI, Multimodality, and Societal Impact

Demis Hassabis, CEO of Google DeepMind, discusses the rapid advancements in AI, emphasizing the imminent arrival of Artificial General Intelligence (AGI) within 5-10 years. He highlights the critical role of multimodal AI, especially in video understanding, and the development of reliable agent-based systems as key short-term developments. Hassabis also addresses the societal implications of AGI, including the need for careful consideration of AI safety, responsible use, and humanity's adaptation to a potentially post-scarcity future.

SIMA 2: A Generalist Embodied Agent Powered by Gemini Achieves Near-Human Performance and Open-Ended Learning in Virtual Worlds

SIMA 2 is an embodied AI agent utilizing a Gemini foundation model, demonstrating advanced interaction capabilities in diverse 3D virtual environments. It surpasses previous iterations by moving beyond simple command execution to engage in goal-directed reasoning, conversation, and multimodal instruction interpretation. This agent exhibits near-human performance in various games and generalizes to novel environments, while also possessing the capacity for autonomous skill acquisition through self-generated tasks and rewards.

Demis Hassabis: The CEO Working to Solve Cancer With AI | Bloomberg Technology

Demis Hassabis on DeepMind's AI Advancements and Future Outlook

Demis Hassabis discusses Google DeepMind's role as the AI engine for Alphabet, integrating advanced models like Gemini across various Google products. He highlights the development of "world models" such as Genie for interactive environment generation, crucial for AGI and robotics. Hassabis also touches upon the application of AI in scientific discovery through Isomorphic, aiming to revolutionize drug discovery and accelerate breakthroughs in fields like material science and health.

Demis Hassabis on World Models, Jagged Intelligence, and the Road to AGI Benchmarks

Google DeepMind is converging its specialized models (Gemini, Genie, Veo) into a unified "omni model" capable of handling multimodal tasks at parity with specialized systems — a trajectory Hassabis frames as necessary for AGI. Current frontier models exhibit "jagged intelligence": superhuman on narrow benchmarks (e.g., 99.2% on AIME, IMO gold medal) yet brittle on simple reasoning tasks, pointing to unresolved gaps in consistency, planning, and memory. To address benchmark saturation and measure progress toward AGI more rigorously, DeepMind is launching Game Arena with Kaggle — a self-scaling, adversarial evaluation environment where model capability determines test difficulty. Genie 3's world model architecture (persistent, physics-consistent world generation) is being used to generate synthetic training data for robotics and general AGI systems, with a Simma agent already operating inside Genie-generated environments.

Hassabis's "Learnable Natural Systems" Conjecture: Classical AI May Model All of Nature's Structured Patterns

In his Nobel Prize lecture, Demis Hassabis proposed that any pattern generated or found in nature can be efficiently discovered and modeled by a classical learning algorithm — a conjecture grounded in the observation that natural systems carry learned structure imposed by evolutionary and physical selection processes. This "survival of the stablest" principle means that proteins, planetary orbits, geological formations, and biological systems all inhabit lower-dimensional manifolds that neural networks can exploit via gradient following. The paradigm is validated empirically by AlphaFold, AlphaGo, and Veo's emergent physics modeling, and Hassabis suggests it points toward a new complexity class — analogous to P and NP — defining problems solvable by neural-network-based classical systems. He views this as a physics question as much as a computer science one, framing information as the most fundamental unit of the universe and P=NP as a core question about the informational structure of reality.

Gemini 2.5 Pro Achieves SoTA on Coding/Reasoning While Spanning Full Capability-Cost Pareto Frontier

Google DeepMind's Gemini 2.X model family introduces a tiered architecture — 2.5 Pro, 2.5 Flash, 2.0 Flash, and Flash-Lite — designed to cover the full capability-vs-cost tradeoff spectrum. Gemini 2.5 Pro is positioned as a "thinking model" achieving state-of-the-art on frontier coding and reasoning benchmarks, with native support for up to 3 hours of video input and long-context multimodal processing. The combination of extended context, multimodal understanding, and reasoning is explicitly framed as an enabler for next-generation agentic workflows. The family's architecture reflects a deliberate design philosophy: match model capability to deployment constraints rather than optimizing for a single frontier point.

Demis Hassabis On The Future of Work in the Age of AI | WIRED

DeepMind CEO Demis Hassabis on How A.I. Is Reshaping Google | Hard Fork Podcast

Google I/O 2024: Shifting AI Strategy and the Road to AGI

Google I/O 2024 revealed a significant shift in the company's AI strategy, emphasizing practical applications and a more confident stance in the AI race. Google is integrating AI across its product ecosystem, notably in Search with "AI mode" and the widespread adoption of Gemini. Despite a focus on product integration, discussions with Demis Hassabis highlight Google DeepMind's continued pursuit of AGI, viewing current advancements as building blocks for future generalized intelligence while acknowledging challenges in productizing rapidly evolving AI capabilities.

Demis Hassabis and Veritasium's Derek Muller talk AI, AlphaFold and human intelligence | The Thinking Game Film

DeepMind CEO Demis Hassabis + Google Co-Founder Sergey Brin: AGI by 2030? - Alex Kantrowitz

MuZero Masters Complex Games via Learned Model Planning Without Dynamics Knowledge

MuZero integrates tree-based search with a learned model to achieve superhuman performance in Atari, Go, chess, and shogi, without prior knowledge of environment dynamics. The model iteratively predicts rewards, action policies, and value functions—quantities essential for planning. On 57 Atari games, it sets a new state-of-the-art; on board games, it matches AlphaZero's superhuman level despite lacking rules.

Machine Learning's High-Impact Applications for Mitigating and Adapting to Climate Change

Machine learning experts can address climate change by applying ML to reduce greenhouse gas emissions and enhance societal adaptation. Key areas include smart grids and disaster management, where ML fills critical gaps through interdisciplinary collaboration. The paper outlines research questions and business opportunities, urging the ML community to prioritize these efforts.

Population-Based RL Achieves Human-Level Play in Multiplayer Quake III Capture the Flag

A population of independent RL agents, trained concurrently across thousands of parallel matches in randomized Quake III Arena Capture the Flag environments, attains human-level performance using only pixel and score inputs. The approach employs a two-tier optimization with self-learned internal rewards supplementing sparse win signals, paired with a temporally hierarchical action representation for multi-timescale reasoning. Agents exhibit human-like behaviors including navigation, following, and defending via encoded high-level game knowledge, outperforming strong humans and prior bots in tournament evaluations.

MERLIN: Predictive Memory Enables RL Agents to Conquer Severe Partial Observability

Standard RL algorithms with deep networks fail on simple tasks under partial observability, even with extensive memory, because they store irrelevant information in suboptimal formats. The MERLIN architecture integrates memory formation guided by predictive modeling, allowing a single agent to maintain long-duration memories and solve partially observable tasks in 3D VR environments. This unifies RL with inference to tackle canonical psychology and neurobiology benchmarks without assumptions on input dimensionality or episode length.

Memory-Based Adaptation Enables Fast, Stable Neural Network Updates Without Catastrophic Forgetting

Memory-based Parameter Adaptation (MemPA) stores training examples in memory and performs context-based lookups to directly modify neural network weights, enabling much higher learning rates than standard gradient updates. This approach accelerates adaptation to distribution shifts, avoids performance degradation on prior data, and mitigates issues like catastrophic forgetting, imbalanced labels, and slow evaluation-time learning. Demonstrated on large-scale image classification and language modeling, it supports fast, stable knowledge acquisition.

State-Space Generative Models Accelerate Model-Based RL with Pixel-Level Atari Dynamics

State-space generative models learn compact representations from raw pixels to predict action sequence outcomes in Atari games, drastically cutting computational costs versus standard models. These models maintain high accuracy on Arcade Learning Environment dynamics. In RL, agents querying these models for planning outperform model-free baselines on Ms. Pac-Man.

Psychlab Enables Psychological Testing of RL Agents, Revealing UNREAL's Size Bias and Foveal Fix

Psychlab integrates classical psychology experiments into DeepMind Lab for testing both human and RL agents via a flexible API, with implementations for visual search, change detection, motion discrimination, and object tracking. Analysis of the UNREAL agent shows it learns faster for larger target stimuli than smaller ones. Adding a foveal vision model corrects this bias and boosts UNREAL's performance on Psychlab and standard DMLab tasks.