DeepMind is Google's AI research lab, known for systems such as AlphaGo. Recent coverage includes Google DeepMind's release of Gemini 4 Argon, a frontier model with a 1M-token context window, and Dream-RSI, which lets agents reuse past search runs to cut iterations by up to 2.43x. Other stories involve alumni: AlphaGo co-creator Thore Graepel founded Metis Reasoning, and Emulate, started by former DeepMind researchers, is reportedly nearing a $700 million seed round.
The long-open van der Waerden number w(2;3,21) is proved to equal 416, with the proof kernel-checked in Lean 4. Kullmann conjectured the value in 2010; the proof spans 209 cases and about 38.7 million leaf formulas, each refutation verified by cake_pr, an LRAT checker verified in CakeML and then bound by digest to its Lean term. It was produced by Hupyy's neuro-symbolic LPM system.
Alphabet shares plunged about 13% after Q3 earnings, wiping out $500 billion in market value. Revenue grew 25% to $127.6 billion but missed estimates, and Search growth slowed to 12%. The author argues Wall Street overlooks Alphabet's vertically integrated AI ecosystem, including custom AI chips, cloud, and DeepMind.
Former DeepMind researcher Thore Graepel founded Metis Reasoning to bring AlphaGo's search principle to general reasoning: rather than locking knowledge in model weights, an explicit knowledge ledger would let every conclusion be checked against evidence. So far there is no working system or benchmark results.
Google engineer Haydn Belfield says DeepMind data shows no measurable acceleration of AI research by AI, and self-improving AI is far off. AI writes over 75% of production code, yet 88% of changes still need human review; the in-house General Capability Index shows steady, not sudden, growth.
The article revisits AI predictions from Kevin Kelly's 2016 book The Inevitable, including that by 2026 Google's main product will be AI not search, that an oligarchy of a few general-purpose cloud intelligences will emerge, that AI will resemble AWS, and that most AI will be special-purpose software brains.
Researchers at the Deepmind Institute propose "Artificial Symbiotic Intelligence," arguing general AI will not emerge as a single supermodel but as a network of cooperating agents and humans. What matters most, they say, is not model size but the rules and institutions governing collaboration.
Google DeepMind released Gemini 4 Argon, a frontier model with a 1M-token context window for complex, long-horizon tasks such as coding, financial research, legal drafting and autonomous vulnerability patching. It is rolling out first to trusted cyber defenders via the Fairwind Program and the US government's pre-release access process. Pricing starts at $2 per million input tokens and $10 per million output tokens, with cached input 95% off.
An Australian tier-2 undergraduate with a first-author NeurIPS 2026 paper is deciding between staying at his current university for a funded PhD or waiting a year to apply to top 15-25 US or UK programs, aiming for a researcher role at DeepMind or Meta.
Thore Graepel, a co-creator of AlphaGo and former Google DeepMind researcher, is raising funds for a new AI reasoning startup. He is the latest prominent researcher to leave DeepMind and seek significant funding.
Google DeepMind's Dream-RSI lets AI agents 'dream' through past search runs to test new strategies without costly recalculations. In tests, it matched or beat existing results, cutting iterations by up to 2.43x, with only the search strategy adapting while the underlying model stays unchanged.
Bloomberg reports that UK AI startup Emulate, founded by former DeepMind researchers, is in talks to raise about $700 million in a seed round, illustrating investor appetite for startups founded by alumni of prominent AI labs. Such a seed round size would be unusually large.
FT reports that a DeepMind co-founder warned AI capabilities must not outpace safety controls, stressing the need to prioritize robust safety and regulatory measures alongside technological progress.
A former DeepMind spokesperson says the lab once prohibited public discussion of AI‑driven human extinction, while internally teams acknowledged that AI alignment remained unsolved.
AI researcher Jacob Coxon resigned from Anthropic, warning superintelligent AI could pose existential risks within the decade; an Anthropic staff lead later estimated the chance could exceed 10%.
That is everything
The A.I. Daily
Van der Waerden number w(2;3,21)=416 proved with AI-assisted Lean proof
