Deepmind veteran David Silver raises $1B seed round to build superintelligence without LLMs

British AI researcher David Silver wants to build "superhuman intelligence" with his new London-based start-up Ineffable Intelligence. The seed round of one billion dollars would be the largest ever raised by a European start-up.

According to a report in the Financial Times, former Deepmind researcher David Silver is raising around one billion dollars for his new start-up Ineffable Intelligence. Sequoia Capital is leading the round, which is expected to value the company at roughly four billion dollars before the new money comes in. According to PitchBook, that would make it the largest seed round ever closed by a European start-up.

Silver left Google Deepmind late last year to launch his new company, also based in London. His departure kicked off a fierce competition among investors, the Financial Times reports: Sequoia managing partner Alfred Lin and partner Sonya Huang flew to London personally to meet with Silver, who also holds a professorship at University College London.

Nvidia, Google, and Microsoft are reportedly in talks to invest as well. Negotiations are still ongoing, and final terms could still shift.

Silver joined Deepmind shortly after it was founded in 2010 and had a hand in some of the AI lab's biggest breakthroughs. He was a key contributor to AlphaGo, AlphaZero, and MuZero. AlphaGo and the related program AlphaStar beat the world's top players in the strategy game Go and the video game StarCraft. After Google acquired Deepmind in 2014, the 49-year-old also played a central role in building Google's Gemini family of language models.

Silver bets on reinforcement learning from experience

Ineffable Intelligence will build on Silver's work in reinforcement learning, an approach where AI systems learn by trial and error instead of training primarily on massive amounts of internet text like most of today's language models. Silver sees LLMs as fundamentally limited because they're built on human knowledge. His goal with Ineffable Intelligence is to build an "endlessly learning superintelligence that self-discovers the foundations of all knowledge."

In a paper on the "Era of Experience" co-authored with computer scientist Richard Sutton, the two predicted that a new generation of agents will acquire superhuman abilities by learning primarily from experience. Experience will become the dominant medium of improvement, they argue, and will ultimately far outpace the amount of human data powering today's systems.

At the core of this vision are so-called world models: internal simulations that let AI agents predict the consequences of their actions. Instead of being trained once and shipped, these agents would continuously adapt to their environment over months or years, much like humans or animals do. AI agents capable of handling complex tasks on their own based on human instructions are increasingly seen as the key to the next phase of AI development and commercialization.

Arthur Mensch previously left Google Deepmind and founded the French start-up Mistral in 2023, raising 105 million euros in its seed round. The Ineffable Intelligence financing would dwarf that European record many times over.

Top researchers increasingly doubt Transformers can get to superintelligence

Silver joins a growing group of top AI researchers who don't think the Transformer architecture alone will get us to superintelligence. Ilya Sutskever, former chief scientist at OpenAI, launched his own start-up, Safe Superintelligence, to explore new approaches beyond current large language models.

Jerry Tworek, who played a key role in building OpenAI's reasoning models, also recently left the company and started Core Automation. Like Silver, he sees continuous learning as one of the last critical pieces before real AI, as long as models can't learn directly from live data, their capabilities hit a ceiling.

Even Deepmind CEO Demis Hassabis has said world models could be the future of AI, predicting a potential "ChatGPT moment," though he sees cost and technical hurdles as the biggest obstacles right now. Hassabis publicly backed his former colleague's move: "I also wish my long time colleague Dave Silver (we've worked together since our student days at Cambridge!) all the best with his exciting new venture! This is how thriving ecosystems are built."