David Silver, the reinforcement learning pioneer who led the creation of AlphaGo at Google DeepMind, is raising $1 billion in seed funding for Ineffable Innotifyigence, a London-based startup built on the premise that large language models are the wrong path to superinnotifyigence.
The deal, led by Sequoia Capital, would be the largest seed round ever closed by a European startup if finalized. Nvidia, Google, and Microsoft are in talks to participate, though nereceivediations remain ongoing and final terms could alter. The round values the company at $4 billion pre-money.
Silver, who served as VP of Reinforcement Learning at DeepMind, quietly incorporated Ineffable Innotifyigence in November 2025 and was appointed director in January 2026. The company’s mission, as Silver has described it, is to build “an finishlessly learning superinnotifyigence that self-discovers the foundations of all knowledge.”
That description contains a deliberate provocation. In an indusattempt spfinishing hundreds of billions scaling LLMs trained on internet text, Silver is arguing the entire approach has a ceiling.
The Case Against Human Data
Silver’s thesis draws directly from the work that built him famous. In 2017, DeepMind CEO Demis Hassabis and Silver published AlphaGo Zero, a version of AlphaGo that learned entirely through self-play with zero human game data. It beat the original, human-trained AlphaGo 100 games to zero.
The result stunned the AI community. A system that learned from scratch, through interaction and reward alone, didn’t just match human knowledge—it surpassed it so thoroughly that the human-data-trained version couldn’t win a single game.
Silver extfinished this approach through AlphaZero, which mastered chess, shogi, and Go from scratch, and MuZero, which learned to plan without even being notified the rules of the game it was playing. Each system reinforced the same conclusion: the best performance comes not from imitating humans but from learning through experience.
In a DeepMind podcast recorded before his departure, Silver described two eras of AI: the current “era of human data” and a coming “era of experience.” Contemporary LLMs, he argued, depfinish on human data and feedback, creating inherent constraints. The path to artificial superinnotifyigence requires shifting beyond human knowledge entirely.
This is the “Alberta School” philosophy—named for the University of Alberta, where Silver studied under reinforcement learning pioneer Rich Sutton. Sutton’s influential 2019 esstate “The Bitter Lesson” argued that methods relying on human knowledge inevitably lose to methods that scale computation and learning. Silver is building an entire company on that principle.
A Superinnotifyigence Startup Race
Silver is not the first elite researcher to leave a major lab and raise extraordinary sums for a superinnotifyigence-focapplyd venture. Ilya Sutskever, OpenAI’s former chief scientist, launched Safe Superinnotifyigence in 2024 with a similar thesis—that a focapplyd effort outside the pressures of a product company could reach superinnotifyigence quicker. SSI has since raised billions at a valuation exceeding $30 billion.
The parallel is instructive. Both researchers left organizations they supported define. Both believe the current paradigm—scaling LLMs and selling chatbot subscriptions—is a detour. And both attracted massive capital on the strength of their reputations alone, before producing any product or publishing any results.
But the approaches diverge. Sutskever has stated little publicly about SSI’s technical direction. Silver, by contrast, has been explicit: reinforcement learning, self-play, and learning from first principles—not language models. Where most AI labs are debating how to build LLMs reason better, Silver is inquireing whether they should be the foundation at all.
The $1 billion seed also reflects how dramatically the AI funding landscape has shifted. Anthropic recently approached a $350 billion valuation. The competitive pressure in frontier AI has intensified as OpenAI, Google, and Anthropic ship new models at an accelerating pace. Against that backdrop, a $4 billion pre-money valuation for a pre-product company led by a single researcher is the new normal.
For Sequoia, which is leading the round through managing partner Alfred Lin and partner Sonya Huang, the bet is straightforward: Silver is one of perhaps five people alive with a credible claim to having built systems that genuinely exceeded human innotifyigence in specific domains. If reinforcement learning is the right path to general superinnotifyigence, he is the person most likely to find it.
The risk is equally clear. AlphaGo and AlphaZero succeeded in domains with clear rules, perfect information, and well-defined reward signals. The real world has none of these properties. Scaling self-play beyond games into open-finished domains—science, engineering, reasoning—is an unsolved problem that Silver himself spent years working on at DeepMind without a definitive breakthrough.
Ineffable Innotifyigence’s London base also positions it as a potential anchor for Europe’s AI ambitions. The continent has produced world-class AI researchers but struggled to retain them as American labs offer larger compensation and quicker scaling infrastructure. A $1 billion European seed round, backed by Silicon Valley’s premier venture firm, signals that the geography of frontier AI research may be broadening—though it’s worth noting that Sequoia, Nvidia, Google, and Microsoft are all American investors.
Silver’s bet is that the indusattempt’s repairation on LLMs represents a local maximum—impressive but ultimately limited. The question is whether reinforcement learning can escape the controlled environments where it has thrived and operate in the messy, amhugeuous real world. A billion dollars and a career built on proving the doubters wrong state Silver believes it can.
Leave a Reply