ClaimsPeople

Demis Hassabis

Sir Demis Hassabis founded DeepMind, leads Isomorphic Labs, shared the 2024 Nobel Prize in Chemistry, and was knighted for services to AI; he was DeepMind's CEO until August 2026, after which he was reported as chairman of Google DeepMind and chief scientist of Alphabet.

802 claims collected, 191 researched, 86 resolved. Report of 4 November 2026, method 0.1.0+c876030c.

Scores by kind of claim

Each kind of claim is scored on its own, out of 100: the share that held up, with the range it likely falls in. Fewer resolved claims give a wider range.

  • Factual claims

    Statements about the present or the past, checked against the record: did they hold up?

    81 76 87

    High confidence, 78 factual claims resolved. From 15 documents, 2004 to 2026; 18 documents sampled from 30 candidates found.

    Not scored: 29 unresolvable, 13 checkable only against their own records.

  • Predictions

    Statements about events outside the person's control, checked once they fell due: did they come true?

    Insufficient record: 2 predictions resolved, 5 needed for a score.

    Not scored: 45 unresolvable. 16 more fall due later.

    Does not yet allow for how likely each prediction was when made.

  • Promises

    Commitments to the person's or organisation's own future action, with something to deliver: was it delivered?

    65 45 83

    Low confidence, 6 promises resolved. From 6 documents, 2004 to 2026; 18 documents sampled from 30 candidates found.

    Not scored: 1 checkable only against their own records.

All types combined: 80 (likely 75 to 85, high confidence), from 84 claims: 78 factual claims, 6 promises. It mixes kinds of claim in different proportions for different people, so compare people kind by kind.

Demis Hassabis's claims

254 claims, sorted by newest said. Choose a claim to read it in full.
VerdictKindSaidClaimTopic
Not researched Factual 23 Jul 2025 DeepMind's AlphaEvolve has incrementally improved specific targets such as faster matrix multiplication. Factual · 23 Jul 2025 AI capabilities
Own records only Factual 23 Jul 2025 Hassabis said it maybe takes roughly six months or something to do a new full run and the full productization of a new version of the model. Factual · 23 Jul 2025 AI capabilities
Not researched Factual 23 Jul 2025 Demis Hassabis used to talk with Paul Nurse, whom he describes as a mentor in biology. Factual · 23 Jul 2025 Their own work or organisation
Not researched Factual 23 Jul 2025 Over the last decade or 15 years, maybe 80-90% of the breakthroughs underpinning the modern AI field came originally from Google Brain, Google Research and DeepMind; he prefaced the figure with "I don't know". Factual · 23 Jul 2025 AI, the economy and work
Not researched Factual 23 Jul 2025 Hassabis has experience working with cutting-edge technology in products, as games were in the 1990s. Factual · 23 Jul 2025 Their own work or organisation
Resolved true Factual 23 Jul 2025 Demis Hassabis worked on the early stages of the game Black & White in the 1990s, which he says had probably the best learning AI in a game. Factual · 23 Jul 2025 Their own work or organisation
Not researched Factual 23 Jul 2025 Demis Hassabis owned a Spectrum and then a Commodore Amiga 500 computer while growing up in the UK in the late 1980s and 1990s. Factual · 23 Jul 2025 Their own work or organisation
Unresolvable Factual 23 Jul 2025 Current AI systems cannot come up with a conjecture worthy of study by a mathematician such as Terence Tao. Factual · 23 Jul 2025 AI capabilities
Not researched Factual 23 Jul 2025 Hassabis said he thinks his own lab is really the only AI lab using AI for science, while other labs talk about science. Factual · 23 Jul 2025 Science and research methods
Unresolvable Factual 23 Jul 2025 Hassabis's bet is that the brain is mostly doing classical computing, which he said would suggest its phenomena can be modelled or mimicked by a classical computer. Factual · 23 Jul 2025 Science and research methods
Mostly true Factual 23 Jul 2025 Video models such as Veo 3 can pretty accurately generate about eight seconds of video that, by eye and at a glance, looks consistent. Factual · 23 Jul 2025 AI capabilities
Own records only Factual 23 Jul 2025 Interim Gemini versions such as 2.5 are often patches or post-training ideas applied to the same base architecture. Factual · 23 Jul 2025 AI capabilities
Not researched Factual 23 Jul 2025 Companies such as Isomorphic Labs were spun out on the back of AlphaFold to do drug discovery, and it is going really well. Factual · 23 Jul 2025 Science and research methods
Not researched Factual 23 Jul 2025 Google DeepMind has models that define the Pareto frontier of performance versus cost and latency. Factual · 23 Jul 2025 AI capabilities
Not researched Factual 23 Jul 2025 Google has its own TPU line and is looking at inference-only chips to make inference more efficient. Factual · 23 Jul 2025 AI compute, hardware and data
Mostly true Factual 23 Jul 2025 Veo, Google DeepMind's video generation model, can model liquids surprisingly well. Factual · 23 Jul 2025 AI capabilities
Not researched Factual 23 Jul 2025 AlphaFold 3 is a first step towards modeling biological interactions, such as proteins with proteins, RNA and DNA. Factual · 23 Jul 2025 Science and research methods
Not researched Factual 23 Jul 2025 Classical computing systems can model protein structures and play Go better than world champion level. Factual · 23 Jul 2025 AI capabilities
Not researched Factual 23 Jul 2025 Hassabis started his journey toward AI more than 30 years ago. Factual · 23 Jul 2025 Their own work or organisation
Not researched Factual 23 Jul 2025 Coding and maths allow large amounts of synthetic data to be created and verified as correct, making them easier for AI to train on. Factual · 23 Jul 2025 AI capabilities
Own records only Factual 23 Jul 2025 After bundling ideas, Google DeepMind starts a new giant 'hero' training run. Factual · 23 Jul 2025 AI capabilities
Resolved true Factual 23 Jul 2025 AI systems are now in products used by billions of people around the world, requiring large amounts of inference compute. Factual · 23 Jul 2025 AI compute, hardware and data
Mostly true Factual 23 Jul 2025 Google DeepMind built DolphinGemma, a version of its system trained on dolphin and whale sounds. Factual · 23 Jul 2025 Science and research methods
Not researched Factual 23 Jul 2025 Google DeepMind has done lots of work with Commonwealth Fusion on AI for fusion reactors. Factual · 23 Jul 2025 Technology
Not researched Factual 23 Jul 2025 Google DeepMind's AlphaProof system achieved a silver medal level on maths Olympiad problems in 2024. Factual · 23 Jul 2025 AI capabilities
Not researched Factual 23 Jul 2025 Demis Hassabis has had the idea of building a Virtual Cell, a model of a cell, for about 25 years. Factual · 23 Jul 2025 Biosecurity
Not researched Factual 23 Jul 2025 As of mid-2025, Meta is not at the frontier of AI research. Factual · 23 Jul 2025 Their own work or organisation
Resolved true Factual 23 Jul 2025 Over the last year a new paradigm of thinking AI systems emerged, which get smarter the more inference time they are given at test time. Factual · 23 Jul 2025 AI capabilities
Own records only Factual 23 Jul 2025 About half of Google DeepMind's resources are devoted to new blue-sky ideas. Factual · 23 Jul 2025 Their own work or organisation
Not researched Factual 23 Jul 2025 Fluid dynamics and Navier-Stokes problems have traditionally been thought very difficult and intractable for classical systems. Factual · 23 Jul 2025 Science and research methods
Unresolvable Factual 23 Jul 2025 Hassabis and the AI community have shown that classical systems (Turing machines) can go much further than previously thought. Factual · 23 Jul 2025 Philosophy, ethics and society
Not researched Factual 23 Jul 2025 When starting DeepMind around 2010, Hassabis paid himself no salary for a couple of years because the company lacked enough money. Factual · 23 Jul 2025 Their own work or organisation
Own records only Factual 23 Jul 2025 Google DeepMind packages research ideas, tests which are useful for the next iteration, and bundles them together for the next model. Factual · 23 Jul 2025 AI capabilities
Not researched Factual 23 Jul 2025 AlphaGo used Monte Carlo Tree Search, which found move 37, a previously unseen strategy in Go. Factual · 23 Jul 2025 AI capabilities
Not researched Factual 23 Jul 2025 Google DeepMind has created the world's best weather prediction systems, which outperform traditional fluid-dynamics systems run on supercomputers. Factual · 23 Jul 2025 Science and research methods
Not researched Factual 23 Jul 2025 John von Neumann reportedly said that computers would be even more impactful in the world. Factual · 23 Jul 2025 Philosophy, ethics and society
Not researched Factual 23 Jul 2025 To Hassabis's knowledge, the neuroscientists who collaborated with the researcher he referred to have not yet found anything convincing on mechanisms for quantum mechanical behaviour in the brain. Factual · 23 Jul 2025 Science and research methods
Mostly true Factual Dec 2024 Hassabis's first career was writing AI for commercial computer games. Factual · Dec 2024 Their own work or organisation
Mostly true Factual Dec 2024 Little progress was made in CASP protein structure prediction accuracy for a decade before AlphaFold. Factual · Dec 2024 Science and research methods
Unresolvable Factual Dec 2024 Levinthal's Paradox estimates about 10^300 possible conformations for a typical protein. Factual · Dec 2024 Science and research methods
Mixed Factual Dec 2024 Hassabis's team has practised open science and open access with all of its scientific work. Factual · Dec 2024 Their own work or organisation
Not researched Factual Dec 2024 Hassabis wrote the game Theme Park when he was 17. Factual · Dec 2024 Their own work or organisation
Not researched Factual Dec 2024 DeepMind consulted with more than 30 biosecurity and bioethics experts to ensure the benefits of AlphaFold outweighed any risks. Factual · Dec 2024 Biosecurity
Not researched Factual Dec 2024 In the 2016 match AlphaGo came up with novel creative strategies, such as Move 37. Factual · Dec 2024 Other
Unresolvable Factual Dec 2024 Hassabis was fascinated as a young chess player by the programming of chess computers used at training camps. Factual · Dec 2024 Their own work or organisation
Not researched Factual Dec 2024 More than 2 million researchers from 190 countries have used AlphaFold, with over 30,000 citations. Factual · Dec 2024 Science and research methods
Not researched Factual Dec 2024 AlphaGo mastered Go through self-learning, playing against itself. Factual · Dec 2024 AI capabilities
Unresolvable Factual Dec 2024 Hassabis ended up wanting to program his own AI systems after encountering chess computers. Factual · Dec 2024 Their own work or organisation
Not researched Factual Dec 2024 Isomorphic Labs was started to build on AlphaFold and reimagine drug discovery from first principles with AI. Factual · Dec 2024 Economics and markets
Not researched Factual Dec 2024 Experimentally determining a single protein structure can often take years of painstaking work. Factual · Dec 2024 Science and research methods
Mostly true Factual Dec 2024 AlphaFold was open sourced and made freely available for the world to use. Factual · Dec 2024 Science and research methods
Mostly true Factual Dec 2024 Proteins are the building blocks of life. Factual · Dec 2024 Science and research methods
Not researched Factual Dec 2024 DeepMind open-sourced AlphaFold and built the AlphaFold database with EMBL-EBI, giving free unrestricted access to every predicted structure. Factual · Dec 2024 Science and research methods
Mostly true Factual Dec 2024 Over two million researchers have used AlphaFold and its predicted structures. Factual · Dec 2024 Science and research methods
Not researched Factual Dec 2024 DeepMind was founded in 2010 as a research lab to build AGI. Factual · Dec 2024 Their own work or organisation
Not researched Factual Dec 2024 The CASP competition has been run biennially since 1994 by John Moult to test computational protein prediction systems. Factual · Dec 2024 Science and research methods
Mostly true Factual Dec 2024 Theme Park had AI at the heart of its gameplay. Factual · Dec 2024 Their own work or organisation
Not researched Factual Dec 2024 AlphaFold1 at CASP13 advanced the field by introducing cutting-edge machine learning as a main component for the first time. Factual · Dec 2024 Science and research methods
Not researched Factual Dec 2024 After decades of experimental work about 170,000 structures had been determined and collated in the Protein Data Bank. Factual · Dec 2024 Science and research methods
Unresolvable Factual Dec 2024 In nature proteins fold spontaneously within milliseconds. Factual · Dec 2024 Science and research methods
Resolved true Factual Dec 2024 In 2016 AlphaGo beat Lee Sedol 4-1 in a match in Seoul. Factual · Dec 2024 Other
Not researched Factual Dec 2024 Demis Hassabis captained all the England junior chess teams while growing up. Factual · Dec 2024 Their own work or organisation
Mostly true Factual Dec 2024 Hassabis first came across the protein folding problem as an undergraduate at Cambridge in the 1990s. Factual · Dec 2024 Their own work or organisation
Not researched Factual Dec 2024 AlphaFold predicted structures for all 200M+ proteins known to science. Factual · Dec 2024 Science and research methods
Mostly true Factual Dec 2024 AlphaFold2 was re-architected for CASP14 to reach atomic accuracy, leading the CASP organisers to declare the problem solved. Factual · Dec 2024 Science and research methods
Not researched Factual Dec 2024 AlphaGo learned from its mistakes to build a useful model of Go. Factual · Dec 2024 AI capabilities
Resolved true Factual Dec 2024 Demis Hassabis learned to play chess at age four. Factual · Dec 2024 Their own work or organisation
Resolved true Factual Dec 2024 AlphaFold2 achieved atomic accuracy at CASP14 in 2020. Factual · Dec 2024 Science and research methods
Unresolvable Factual 21 Oct 2024 Around 15 years ago, when DeepMind was founded, almost nobody was working on artificial general intelligence/AI of this kind. Factual · 21 Oct 2024 Their own work or organisation
Unresolvable Factual 28 Feb 2024 Before founding DeepMind, its founders already knew that progress on these ideas would create unbelievably transformative technology. Factual · 28 Feb 2024 Their own work or organisation
Not researched Factual 28 Feb 2024 DeepMind's pioneering work, inspired by neuroscience, included experience replay and the notion of attention. Factual · 28 Feb 2024 Their own work or organisation
Not researched Factual 28 Feb 2024 In 2010, no one was thinking about AI, let alone AGI. Factual · 28 Feb 2024 Their own work or organisation
Own records only Factual 28 Feb 2024 Google DeepMind devotes about half its effort to scaling and half to inventing next architectures and algorithms. Factual · 28 Feb 2024 Their own work or organisation
Own records only Factual 28 Feb 2024 Hassabis said that "we" have a lot of processes in place to make sure that only people who need access have access. Factual · 28 Feb 2024 AI safety and alignment
Mostly true Factual 28 Feb 2024 DeepMind has always bet on generality and learning as the core of its techniques, focusing on reinforcement learning, search and deep learning. Factual · 28 Feb 2024 Their own work or organisation
Mostly true Factual 28 Feb 2024 Demis Hassabis thinks human brains plan moves differently from AlphaGo because brains are not built for Monte Carlo tree search. Factual · 28 Feb 2024 Science and research methods
Research failed Factual 28 Feb 2024 Early signs show large models becoming transferable to robotics. Factual · 28 Feb 2024 AI capabilities
Not researched Factual 28 Feb 2024 Since DeepMind's founding, it has said that it does not have to wait for AGI to bring incredible benefits to the world. Factual · 28 Feb 2024 Their own work or organisation
Mostly true Factual 28 Feb 2024 Demis Hassabis thinks that improving in one specific domain sometimes produces surprising improvement in other domains; as an example, he said that when large models improve at coding, that can improve their general reasoning. Factual · 28 Feb 2024 AI capabilities
Not researched Factual 28 Feb 2024 DeepMind joined forces with Google in 2014. Factual · 28 Feb 2024 Their own work or organisation
Unresolvable Factual 28 Feb 2024 Google DeepMind has covered the secure-model requirement through world-class cybersecurity and monitoring. Factual · 28 Feb 2024 AI safety and alignment
Not researched Factual 28 Feb 2024 When large AI models improve at coding, their general reasoning can also improve. Factual · 28 Feb 2024 AI capabilities
Not researched Factual 28 Feb 2024 RL agents like AlphaZero and AlphaGo are active learners whose decisions determine what data or experience they get next. Factual · 28 Feb 2024 AI capabilities
Not researched Factual 28 Feb 2024 Downstream capabilities such as MMLU or math performance sometimes don't follow from predictable improvements in training loss. Factual · 28 Feb 2024 Science and research methods
Unresolvable Factual 28 Feb 2024 Google DeepMind operates behind Google's firewall and cloud protection, which Hassabis said he thinks is best in class corporately. Factual · 28 Feb 2024 AI safety and alignment
Mostly true Factual 28 Feb 2024 DeepMind and Google Brain previously collaborated project by project, and now collaborate more deeply and broadly. Factual · 28 Feb 2024 Their own work or organisation
Mostly true Factual 28 Feb 2024 Gemini is the first product of the DeepMind–Google Brain collaboration, and its name implies twins. Factual · 28 Feb 2024 Their own work or organisation
Not researched Factual 28 Feb 2024 Google DeepMind is still carrying out robotics research. Factual · 28 Feb 2024 Their own work or organisation
Mixed Factual 28 Feb 2024 Large models of the Gemini and GPT-4 generation generalize to new constructs rather than merely memorizing. Factual · 28 Feb 2024 AI capabilities
Resolved true Factual 28 Feb 2024 Hassabis said he thinks Google DeepMind helped pioneer RLHF. Factual · 28 Feb 2024 AI capabilities
Own records only Factual 28 Feb 2024 Hassabis said he thinks Gemini 1 used roughly the same amount of compute as was rumored for GPT-4, maybe slightly more, though he said he did not know exactly what was used for GPT-4. Factual · 28 Feb 2024 AI compute, hardware and data
Not researched Factual 28 Feb 2024 Google DeepMind and Google have published thousands of papers, including AlphaFold, transformers and AlphaGo. Factual · 28 Feb 2024 Their own work or organisation
Not researched Factual 28 Feb 2024 AlphaZero and AlphaGo may look at around tens of thousands of possible positions to decide their next move. Factual · 28 Feb 2024 AI capabilities
Not researched Factual 28 Feb 2024 Scaling laws require scaling up hyperparameters and new innovations at each new scale, not just repeating the same recipe. Factual · 28 Feb 2024 Science and research methods
Resolved true Factual 20 Apr 2023 Zoubin Ghahramani will join the Google DeepMind research leadership team reporting to Koray Kavukcuoglu. Factual · 20 Apr 2023 Their own work or organisation
Resolved true Factual 20 Apr 2023 Google DeepMind brings together DeepMind and Brain AI talent with Google's computing power, infrastructure and resources. Factual · 20 Apr 2023 Their own work or organisation
Resolved true Factual 20 Apr 2023 On 20 April 2023 Sundar Pichai announced that DeepMind and Google Research's Brain team would join as a single unit called Google DeepMind. Factual · 20 Apr 2023 Their own work or organisation
Mostly true Factual 20 Apr 2023 Eli Collins will join Demis Hassabis's leads team at Google DeepMind as VP of Product. Factual · 20 Apr 2023 Field building and careers
Mostly true Factual 20 Apr 2023 When DeepMind launched in 2010, many people thought general AI was farfetched science fiction decades away. Factual · 20 Apr 2023 Their own work or organisation
Mostly true Factual 20 Apr 2023 Research advances from the Brain and DeepMind teams, from Deep Reinforcement Learning to Transformers, laid much of the foundations of the current AI industry. Factual · 20 Apr 2023 Their own work or organisation
Back to the list

Factual AI capabilities

“And as you know with AlphaVol, like things like faster matrix multiplication, so when you hone it down to very specific thing you want, it’s very good at incrementally improving that.”

Demis Hassabis, CEO of Google DeepMind, interviewee 23 July 2025 Transcript for Demis Hassabis: Future of AI, Simulating Reality, Physics and Video Games (#475) (Lex Fridman)

DeepMind's AlphaEvolve has incrementally improved specific targets such as faster matrix multiplication.

Not researched Checkable, but outside the sample that was researched.

Said for an organisation they lead or work for; it counts on both records.

Cite this claim

Demis Hassabis, “DeepMind's AlphaEvolve has incrementally improved specific targets such as faster matrix multiplication.” (said 23 July 2025). Verdict: not researched. Reliability API, claim c_25016ae12dd7, method 0.1.0+c876030c. https://epistacklabs.com/claims/c_25016ae12dd7

Claims without a verdict

Is a verdict here wrong, or do you speak for Demis Hassabis? Reply on the claim's own page, with the form at its foot (how replies work). Verdicts are made automatically; how verdicts and scores are made.