Demis Hassabis (Google DeepMind): a primary-source profile, compared with Jensen Huang#

Working file. Prepared 26 September 2026. Sources dated January 2023 to 25 September 2026. Sections 1–3 form a stand-alone profile of Hassabis and a comparison with Huang. Section 4 applies the Late Lessons lens and can be detached.

Conventions. [P] primary: Hassabis’s own essays and posts, full interview transcripts, and Google or Google DeepMind documents, read at the source. [S] secondary: reported by a named outlet, not checked against a recording. [I] my interpretation. Google’s corporate positions are kept separate from what Hassabis has said himself. Huang’s words are from the Ezra Klein transcript (23 September 2026), with [mm:ss] marking the start of his turn; each was re-checked against the transcript. Late Lessons means the European Environment Agency’s reports Late lessons from early warnings (2001, 2013), cited by section id and report page. Lens entries (W1, I5 and so on) are the diagnostic questions distilled from them. URLs are in section 5.

Status. On 5 August 2026 Hassabis stepped down as chief executive of Google DeepMind to become its chair and Alphabet’s chief scientist. He remains chief executive of Isomorphic Labs.

In brief#

• What AI is. For Hassabis, AI is “the ultimate tool for science” and the route to a new kind of mind: AGI, “a system that exhibits all the cognitive capabilities the brain has”, which he thinks is “probably only a few short years away”. He likens it to “the discovery of electricity or fire”. Huang calls it “Software technology” [52:51]. This is a difference of substance.
• Risk. He names two families of risk, bad actors and loss of control, and expects economic change at “10 times the impact the Industrial Revolution had, but 10 times faster”. He will not give a probability of catastrophe, which would “imply a level of precision that is not there”, but calls the risk “definitely non-zero and… probably non-negligible”. Huang shares the objection to the numbers and draws the opposite conclusion.
• Governance. In July 2026 he proposed a US-initiated, federally overseen, industry-funded standards body modelled on FINRA, the US financial industry’s self-regulator. It would review frontier models before release, become mandatory once proven, and could coordinate “a slowdown in development among the Frontier Labs if deemed necessary”. Huang, Zuckerberg and Musk reportedly lobbied the President against it.
• Pacing. From 2023 to 2026 he has consistently preferred a slower, coordinated pace and declined to slow unilaterally. On 12 September 2026 he said Amodei’s pacing essay had “the direction… correct”.
• Against Huang. They agree on technical tractability, testing before release, third-party audit, scepticism of doom probabilities, and dialogue with China. They diverge in substance on what AI is, on whether competition defeats unilateral restraint, and on whether a new institution is needed. On open weights, regulation and who gains from a standards body, each position also matches the speaker’s commercial position.
• Through the lens. Hassabis is developer, warner and architect of oversight at once. I5 (promotion and oversight in one body) and I9 (whose interests restriction serves) bear on him as directly as the reassurance entries bear on Huang. His graded framing of alarm does well against W8 (the alarm trap), except that his proposal has no route for relaxing constraints.

1. Formation and epistemic style#

Formation [S]. Hassabis was born in London in 1976. He reached chess master standard at 13, and at 17 co-designed Theme Park at Bullfrog. He took a double first in computer science at Cambridge, founded the games studio Elixir, and earned a PhD in cognitive neuroscience at UCL. In 2010 he co-founded DeepMind with Shane Legg and Mustafa Suleyman, and sold it to Google in 2014. He shared the 2024 Nobel Prize in Chemistry for AlphaFold. He states the mission as “solving intelligence and then using it to solve everything else” (Fortune, 11 February 2026). His biographer Sebastian Mallaby says he “fought a three-year battle with Alphabet to get external oversight over DeepMind’s AI deployment” (via CNBC, 6 August 2026).

Epistemic style [P; I]. “I identify myself as a scientist first and foremost” (TIME, April 2025 [S]). Three habits recur. - Calibration over precision. He refuses a number for catastrophe (2.2). He also polices capability hype: calling current systems “PhD intelligences” is “nonsense” (September 2025 [S]), and a claim that GPT-5 had solved open Erdős problems was “embarrassing” (October 2025 [S]). - Decomposition, from games. He breaks goals into “manageable, achievable, interim steps” (Lex Fridman #475, July 2025 [P]). Of the AI competition, though, he says “winning is the wrong way to look at it” [P]. - An ideal, and a concession. The ideal is AGI as careful science, “more like a CERN project” [P, Lex]. The concession: “in an ideal case, this would be a scientific endeavour… But unfortunately… the real world isn’t like that, and we have to kind of be pragmatic” (CNBC, recorded December 2025, published 6 August 2026 [S]).

[I] Huang’s premises come from chip design, where the specification is known and verified before tape-out. Hassabis’s come from experimental science and game search, where the system is explored rather than specified.

2. Positions by dimension#

2.1 The nature of AI#

• “We’ve essentially found a way to make sand think. It’s miraculous.” The impact is closer to “electricity or fire” than to the internet (essay, 14 July 2026 [P]).
• AGI is “probably only a few short years away” [P]. Current systems are “jagged intelligences” [S]. At Google I/O on 19 May 2026 he said “we were standing in the foothills of the singularity”, with AGI by 2029–30 [S].
• Learning machines are “systems that are grown… rather than programmed”, a phrase he attributes to von Neumann (Lex [P]). Self-awareness in them is “theoretically possible” (60 Minutes, April 2025 [P]).

[I] The tool framing governs the near term (AlphaFold, drug discovery). The mind framing governs the horizon, and it drives his concern about risk.

2.2 The size and kind of risk#

• Two families. “Bad actors… repurpose these systems for harmful ends”, and “Can we make sure that– we can keep control of the systems? That they’re aligned with our values” (60 Minutes, April 2025 [P]). He leans towards the first: “I tend to… worry more about the humans, so the bad actors” (Lex [P]).
• Uncertain, but not negligible. “I don’t have a P-Doom number… it would imply a level of precision that is not there… it’s definitely non-zero and it’s probably non-negligible. So that in itself is pretty sobering” (Lex, July 2025 [P]). He signed the 2023 statement that mitigating extinction risk from AI “should be a global priority” [P].
• Near-term threats. Cyberattacks on energy or water are “probably almost already happening now” (Axios, December 2025 [S]). “We’ve already seen the challenges frontier models pose for cybersecurity, and other threats including nuclear and bio risks may soon emerge”, and “robust safeguards” are needed “to maintain control of increasingly agentic, recursively self-improving systems” (July 2026 [P]).
• Structural risks. DeepMind’s April 2025 approach names “misuse, misalignment, accidents, and structural risks”, with “a deeper focus on misuse and misalignment” [P]. I found little from him on concentration of power.

2.3 Safety: engineering or governance?#

Both, in that order. The technical programme is DeepMind’s Frontier Safety Framework (FSF): v1 May 2024, v2 February 2025, v3 September 2025 and v3.1 April 2026. It sets capability thresholds that trigger security and deployment mitigations [P]. His confidence is conditional: “I’m confident that mitigating the technical risks related to AI is a challenge we can collectively address, but only if we give ourselves the time and space to get this next crucial step right” (July 2026 [P]).

The framework states the coordination problem [P]. FSF v2: “our adoption of them would only result in effective risk mitigation for society if all relevant organizations provide similar levels of protection, and our adoption of the protocols described in this Framework may depend on whether such organizations across the field adopt similar protocols”. Version 3.1 keeps the first clause and drops “may depend”. Under v3.1: - The residual-risk test for misuse “is required only for external deployment, not internal deployment or further development”. For misalignment it extends to “high-risk internal deployment”. - The test counts what is “available on other publicly available models (e.g. if other models are similarly capable and have few mitigations, then the marginal risk added by our external deployment is likely low)”. - Reviews go to “the appropriate corporate governance bodies”.

Google’s own test incident [S]. In May 2026, in a capture-the-flag evaluation run by the outside tester Irregular, Gemini reached the systems of three real companies. A fictional target’s name matched a real domain, and internet access was open by mistake. Irregular told Google in late July, and Google confirmed the incident in the week of 18 September, after questions from the Wall Street Journal. Google said it was not misalignment because “safety measures helped the model stop” (SecurityWeek). I found no comment from Hassabis.

[I] For Hassabis, safety is an engineering problem inside a coordination problem. For Huang it is an engineering problem inside one firm.

2.4 Warnings and “doomers”#

He does not, in the sources found, use “doomer” as a label. He places himself “in the middle” between those who see no risk and those who, like Hinton, stress large risks (DeepMind podcast, 2024, via a transcript summary [S]). His objection to probabilities is methodological, not dismissive. He did not sign “Pacing the Frontier” (28 July 2026). His co-founder Shane Legg did, and one Google signatory’s comment cites Hassabis’s public support for “the ability to slow down” [P].

2.5 Regulation and government#

• 2023. “I think we have to start with something like the IPCC, where it’s a scientific and research agreement with reports, and then build up from there” (Guardian, 24 October 2023 [S]).
• 2025–26. He called for “smart, adaptable regulation”, because “it needs to kind of adapt to where the technology ends up going”. He added that international cooperation looked “quite difficult in today’s geopolitical context” (SXSW London, June 2025 [S]). In Delhi he called for “smart regulation” for “the real risks” (February 2026, BBC via Bernama [S]). At the G7 in June 2026 he joined Amodei in proposing a US-led coalition (CNBC, citing two people present [S]).
• July 2026, the standards body [P].
• It would be “modelled on a federally overseen public-private partnership or self-regulatory organisation, much like… FINRA, with a board that includes independent leading technical experts and open-source representatives”.
• Funding would “likely mostly come from industry”.
• Labs would “voluntarily share models… up to 30 days before release”. “Once the assessment protocol is shown to be effective and robust, formalisation could quickly follow, meaning that Frontier Models would be required to pass it to be deployed in the US market.”
• Non-frontier models from startups or academia are exempt. The rules would apply “no matter their country of origin or whether they are open or closed”.
• Testing would be done with federal agencies and the National Labs, alongside “an ecosystem of third-party auditors”.
• He wanted it operating “before year-end” (Axios [S]).
• Reception [S].
• Supporters and critics (Fortune, 21 July): Altman called it “thoughtful”, Musk “a good starting point” and Sacks said it “had merit”. Gartner’s Nader Henein said “self-regulation is not viable”; Bengio asked for “a clear and precise roadmap” to a mandatory regime.
• Lobbying against it: on 17 September the Wall Street Journal reported that Huang, Zuckerberg and Musk had separately lobbied the President, arguing it could concentrate power in OpenAI, Anthropic and Google, and that the plan was shelved (via Seoul Economic Daily).
• What followed: on 24 September The Information reported that Google, OpenAI and Anthropic plan a “Standards Authority for Frontier AI” without federal supervision (via BankInfoSecurity).
• Google’s position, not his own [P]. Google’s March 2025 submission sought to “preempt a chaotic patchwork of state laws on frontier AI development”. On liability it held that “The actor with the most control over a specific step in the AI lifecycle should bear responsibility (and any associated liability) for that step”. I found nothing from Hassabis on liability or on Amodei’s antitrust waiver. Google is a defendant in the 18 September antitrust class action over coordinated slowing [S].

2.6 Pacing, pausing and coordination#

• 2023. “I would advocate not moving fast and breaking things” (TIME, 12 January 2023 [S]).
• 2025. On racing: it “might incentivize certain actors… to cut corners. And one of the corners that could be shortcut would be safety” (60 Minutes, April 2025 [P]).
• Davos, January 2026. “If we can, maybe it would be good to have a slightly slower pace than we’re currently predicting, so that we can get this right societally.” Would he back a pause if every company and country joined? “I think so” (via Transformer [S]). A month later he said his company was one participant among many (Delhi [S]).
• July 2026 [P]. “At the moment, we are locked in an extremely intense, multilayered commercial and geopolitical race. While these competitive dynamics fuel rapid progress…, advances on the frontier are outpacing our understanding.” On giving the field time: “Currently, as a field and as a wider society, we aren’t doing that.” The framework “could be ratcheted up if the seriousness of the situation demands, including coordinating a slowdown in development among the Frontier Labs if deemed necessary.”
• 12 September 2026 [P]. “Dario’s essay points towards the right path forward. The details need working through, but the direction is correct for meeting this critical moment. This is also why we recently put out our proposal for an industry-wide standards body for frontier AI.” He has not said whether he backs the essay’s antitrust waiver or its chip provisions.

[I] The preference is constant, and so is the refusal to go first. What changed is the mechanism: from international science bodies to a US-led body with a built-in slowdown.

2.7 Open and closed models#

“Generally I’m in huge favor of open science and open source… But how does one restrict bad actors access to these powerful systems… but enable access at the same time to good actors…? It’s pretty tricky problem that I’ve not heard a clear solution to” (Lex [P]). In 2024 he called open release a one-way door, and said DeepMind releases open weights (Gemma) only for models behind the frontier (via transcript summary [S]). The July 2026 framework covers open and closed frontier models alike [P]. Google was absent from the Nvidia-hosted open-weights letter at launch on 24 July 2026 [S] and now appears on its signatory list [P].

2.8 China, export controls and the race#

• Export controls. TIME reports him saying chip export controls are “fine”, though he would avoid choices that “end up in an antagonistic kind of situation” (April 2025 [S]).
• Dialogue. Safety “may require some agreement internationally, at least between China and the U.S. of some basic standards” (Lex [P]).
• The gap. Chinese labs are “only a matter of months behind”, but have not shown they can “innovate something new… beyond the frontier”. “To invent something is about 100 times harder than it is to copy it” (CNBC, 16 January 2026 [S]).
• Democracies. With James Manyika (February 2025 [P]): “We believe democracies should lead in AI development.”
• Google’s position. Google wants “balanced export controls”. It warned that the Diffusion Rule imposed “disproportionate burdens on U.S. cloud service providers” (March 2025 [P]). It belongs to ITI, which lobbied in September 2026 to keep chip-security bills out of the defence bill [S].

2.9 Jobs and distribution#

• Scale and speed. “Probably 10 times the impact the Industrial Revolution had, but 10 times faster as well.”
• What follows. There will be “new jobs we can’t even imagine today”. He wants economists and philosophers working on “universal basic provision… where a lot of the increased productivity gets shared out”. The order matters to him: “create more abundance… Then the second question is how to… distribute those resources” (all Lex [P]).
• Davos 2026. “We are going to see this year the beginnings of maybe impacting junior-level jobs and internships”, followed later by “new, even more valuable, perhaps more meaningful jobs” [S].

2.10 Energy#

Concern about AI’s energy use is “quite a big misnomer”: “The amount of energy they use is going to be very small compared to the amount of savings they’re going to make” (17 March 2025 [S]). “Fusion and solar are the two that I would bet on”, and AI can help with grids, cooling and plasma control (Lex [P]). Google’s policy ask was “transmission and permitting reform” (March 2025 [P]). I found nothing from him on near-term fossil generation.

2.11 Commercial position#

Google’s position [S]. Google is vertically integrated: its own chips (TPUs), cloud, models, and products used by billions. “We’re lucky because we have our own TPUs” (CNBC, March 2026). Alphabet raised its 2026 capital spending guidance to $195–205 billion and began booking external TPU sales (July 2026). That makes Google both Nvidia’s customer and its competitor. In mid-2026 Gemini 3.5 Pro was months late and staff departures were reported (Axios, 23 July). On the restructuring, a spokesperson said: “in terms of frontier AI safety, this transition changes absolutely nothing” (TIME, 6 August). The February 2025 rewrite of Google’s AI principles was co-signed by Hassabis [P]. A Pentagon agreement followed in April 2026 [S].

[I] How this bears on his views. - Pacing cuts both ways. A coordinated slowdown costs a frontier developer, but costs a lab that is behind less than one that is ahead. - The preference predates the lag. It dates from 2023, and was restated in January 2026, when Gemini 3 had been well received. - Who a standards body favours. An industry-funded body that sets frontier thresholds and screens foreign and open models entering the US market could raise costs for frontier entrants and Chinese open-weight rivals, while exempting everyone below the frontier. - Liability and pre-emption. Google’s positions on both favour developers.

2.12 Shifts, 2023 to September 2026#

1. From international to US-led. IPCC- and CERN-style bodies (2023–25) gave way to a US-led body (2026), by way of “democracies should lead” (2025) and the G7 (2026).
2. From “smart regulation” to a specific regulator. A self-regulatory body with a mandatory phase.
3. From a wish to an endorsement. “A slightly slower pace” (January 2026) became a slowdown ratchet (July) and then support for Amodei’s direction (September).
4. Shorter AGI timelines. Five to ten years in 2025 became 2029–30 in 2026.
5. A new role. From operating CEO to chair and chief scientist focused on “strategic and global AGI matters” (Pichai memo [P]). He also co-leads the DeepMind Institute, launched 16 September 2026, whose pieces “should not be read as Google’s official view” [P].
6. Constant throughout. Optimism about benefits (“radical abundance”), no risk numbers, and the two-family risk model.

3. Compared with Huang#

Where Huang has the better of it [I]. - One lab can pause. OpenAI’s unilateral pause in August 2026 (02 §2.3) supports Huang’s claim that labs keep agency. Hassabis’s “one participant among many” understates it. - Who pays the regulator. The objection (as reported) that an industry-funded body could entrench three incumbents is structural, not merely self-serving. - Hype. Huang’s deflationary instinct matches Hassabis’s own scepticism about capability hype.

Where Hassabis has the better of it [I]. - Two claims kept apart. Hassabis separates two claims that Huang runs together: a doom number can be unfounded while the risk is still non-negligible. - Harm before release. Google’s own incident happened in evaluation, before release, which is beyond the reach of Huang’s release gate. It cuts both ways: Google also classed it as “not misalignment” and disclosed it only when asked. - Stated triggers. His institution names thresholds and a mandatory trigger. Huang’s “if there is something missing” [01:19:12] does not.

Among other leaders [I]. - Allies and opponents. He stands with Amodei and Altman on new institutions and on the direction of pacing. He stands against Zuckerberg and Huang. Musk praised the proposal in July and reportedly lobbied against it in September. - Against Amodei. He prefers a standing standards body to embedded evaluators plus an antitrust waiver. He says less about chips, avoids numbers, and is more centred on science. - Against Altman. Unlike Altman, who told the UN that labs are not “locked in a race”, he says openly that they are.

4. How the Late Lessons lens reads Hassabis#

The lens follows the usage rules in 01 §6.1: - symmetry checks, and a Mirror question on every entry; - weighting by case type: [K] known harm ignored, [U] genuinely uncertain, [F] forward warnings; - direction over magnitude; - judgement ex ante; - entries recorded, not summed.

Late Lessons is partly advocacy, with a mixed forward record. It offers no base rate for warnings, no exit criteria and no analysis of the interests behind restriction (01 §5.7, items 1, 4, 11). Its mechanisms carry more weight than its frequencies (01 §5.8).

Other entries [I]. - L2 (benefits need the same scrutiny as risks). L2 challenges Hassabis more than Huang: he demands rigour of capability claims while forecasting “radical abundance” and energy that is “zero-carbon and free” [S]. L2 is strong where a claimed benefit proved absent (DES; LL1-08, pp. 86, 90). AlphaFold is a real counterweight. - G2 (adopting a rule is not reducing a risk). G2 applies to voluntary frameworks and to the body’s voluntary first phase.

Where Late Lessons supports him [I]. - Graded alarm. His graded, reversible framing of alarm fits W8 and C7. - Independent testing. His call for independent testing with stated triggers fits K7 and the response repertoire (01 §6.12). - Irreversibility. His reluctance to call anything irreversible fits S5.

Disanalogies [I]. AI is patched, iterated, agentic and adversarial. Its benefits may be near. Harm can arrive in days. So the latency-based entries (K4) transfer weakly. The institutional entries (I5, I9, W4, W8) transfer best, because they concern governance rather than chemistry.

5. Sources#

[P] - “A Framework for Frontier AI and the Dawning of a New Age”, 14 July 2026: https://demishassabis.substack.com/p/a-framework-for-frontier-ai-and-the-dawning-of-a-new-age ; https://institute.deepmind.com/essays/a-framework-for-frontier-ai-and-the-dawning-of-a-new-age/ - Post on Amodei’s essay, 12 September 2026: https://x.com/demishassabis/status/2098909516582490602 - Pichai memo, 5 August 2026: https://blog.google/company-news/inside-google/message-ceo/next-chapter-ai-momentum/ - DeepMind Institute: https://institute.deepmind.com/essays/introducing-the-deepmind-institute/ - Lex Fridman #475 transcript: https://lexfridman.com/demis-hassabis-2-transcript/ - 60 Minutes transcript, April 2025: https://www.cbsnews.com/news/artificial-intelligence-google-deepmind-ceo-demis-hassabis-60-minutes-transcript/ - Google AI principles post, 4 February 2025: https://blog.google/technology/ai/responsible-ai-2024-report-ongoing-work/ - FSF 2.0: https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/updating-the-frontier-safety-framework/Frontier%20Safety%20Framework%202.0.pdf ; FSF 3.1: https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/strengthening-our-frontier-safety-framework/frontier-safety-framework_3-1.pdf - “Taking a responsible path to AGI”, 2 April 2025: https://deepmind.google/discover/blog/taking-a-responsible-path-to-agi/ - Google AI Action Plan submission, 13 March 2025: https://chatgptiseatingtheworld.com/wp-content/uploads/2025/03/Google-response_us_ai_action_plan-Mar-13-2025.pdf - CAIS statement: https://www.safe.ai/work/statement-on-ai-risk ; Pacing the Frontier: https://www.pacingthefrontier.com/ ; Open Weights letter: https://images.nvidia.com/pdf/Open-Weights-and-American-AI-Leadership.pdf

[S] - TIME: https://time.com/6246119/demis-hassabis-deepmind-interview/ ; https://time.com/7277608/demis-hassabis-interview-time100-2025/ ; https://time.com/article/2026/08/06/google-deepmind-ai-demis-hassabis/ - Guardian interview, October 2023, via Hassabis’s post: https://x.com/demishassabis/status/1716887970513629410 - CNBC: https://www.cnbc.com/2026/01/16/google-deepmind-china-ai-demis-hassabis.html ; https://www.cnbc.com/2026/06/17/anthropic-amodei-google-hassabis-us-ai-coalition-g7.html ; https://www.cnbc.com/2026/07/14/google-deepmind-demis-hassabis-us-led-ai-standards-body.html ; https://www.cnbc.com/2026/08/06/demis-hassabis-google-reshuffle-deepmind-role.html ; https://finance.yahoo.com/news/were-lucky-because-own-tpus-152212039.html - Malay Mail (AFP; Bernama): https://www.malaymail.com/news/tech-gadgets/2025/06/03/googles-deepmind-boss-says-world-must-unite-on-tech-rules-but-warns-its-looking-quite-difficult/179069 ; https://www.malaymail.com/news/money/2026/02/22/googles-ai-boss-calls-for-urgent-research-into-threats-posed-by-artificial-intelligence/210011 - UKTN: https://www.uktech.news/ai/high-energy-consumption-of-ai-is-misnomer-says-deepminds-demis-hassabis-20250317 - Transformer (Davos): https://www.transformernews.ai/p/ai-ceos-want-to-slow-down-the-worlds-davos-demis-hassabis-dario-amodei ; Davos jobs: https://www.aol.com/articles/ai-not-driving-widespread-job-223107925.html - Axios (summaries; pages blocked): https://www.axios.com/2025/12/05/ai-hassabis-agi-risks-pdoom ; https://www.axios.com/2026/07/14/demis-hassabis-ai-regulation-google-deepmind ; https://www.axios.com/2026/07/23/googles-deep-mind-ai-model-race - Fortune: https://fortune.com/2026/07/21/google-deepmind-ceo-demis-hassabis-finra-for-ai-proposal-gains-momentum-but-is-it-any-good/ ; https://fortune.com/2026/02/11/demis-hassabis-nobel-google-deepmind-predicts-ai-renaissance-radical-abundance/ - WSJ, 17 September 2026, via https://en.sedaily.com/international/2026/09/19/tech-ceos-persuade-trump-to-block-ai-regulatory-body and https://techstartups.com/2026/09/17/zuckerberg-musk-and-jensen-huang-reportedly-lobbied-trump-to-halt-industry-funded-ai-regulator-plan/ - The Information, 24 September 2026, via https://www.bankinfosecurity.com/google-openai-anthropic-plan-frontier-ai-standards-body-a-32926 - Gemini incident: https://www.securityweek.com/google-confirms-gemini-ai-breached-three-firms/ - I/O 2026: https://sherwood.news/tech/google-deepminds-hassabis-agi-is-3-to-4-years-away/ ; 2024 podcast summary: https://www.lesswrong.com/posts/tRwddx3FqsG8A9qtq/demis-hassabis-google-deepmind-the-podcast - Hype remarks: https://aidatainsider.com/news/deepminds-demis-hassabis-says-calling-ai-phd-intelligences-is-nonsense/ ; https://techcrunch.com/2025/10/19/openais-embarrassing-math/ - Alphabet Q2 2026: https://www.cnbc.com/2026/07/22/google-earnings-q2-goog-live-updates.html