Google Deepmind pioneer David Silver departs to found AI startup, betting LLMs alone won't reach superintelligence

Google Deepmind loses one of its leading AI researchers as David Silver departs, adding to growing skepticism that current language models alone can achieve superintelligence.

Silver was one of the AI lab's first employees and played a central role in landmark projects like AlphaGo, AlphaZero, and MuZero. According to Fortune, Silver has left to start his own company in London called Ineffable Intelligence. The startup was registered in November 2025 and is currently seeking investors and AI researchers.

A Google Deepmind spokesperson confirmed Silver's departure to Fortune and called his contributions "invaluable." Silver's focus is now superintelligence, AI that surpasses humans in every meaningful way. He sees large language models as fundamentally limited because they're built on human knowledge. Instead, he's betting on reinforcement learning, where AI learns through trial and error.

It's worth noting that Deepmind isn't opposed to superintelligence. In fact, it's been Deepmind founder Demis Hassabis's lifelong goal. The difference likely lies in how to get there. And as a recent discussion between Deepmind's Adam Brown and former Meta AI leader Yann LeCun showed, some researchers at Deepmind still believe LLMs can go a long way, as do many at Anthropic and OpenAI.

Silver's vision: AI that learns from experience, not human data

In April 2025, Silver co-authored a paper with renowned AI researcher Richard Sutton calling for a fundamental shift in AI development, away from training on human knowledge and toward systems that learn from their own experience.

The authors call this the "Era of Experience" and center their vision on world models, simulations that let AI agents predict the consequences of their actions. A key piece is continuous learning: rather than being trained once, AI agents would adapt to their environment over months or years, much like humans or animals. Silver says Ineffable Intelligence aims to build an "endlessly learning superintelligence that self-discovers the foundations of all knowledge."

Growing skepticism about current AI approaches

Silver joins a growing number of leading AI researchers who doubt the current Transformer architecture can achieve superintelligence, or even go much beyond what it's currently capable of, though some headroom remains. Ilya Sutskever, former chief scientist at OpenAI, founded his own startup called Safe Superintelligence to explore new approaches beyond large language models.

Jerry Tworek, who helped develop OpenAI's reasoning models, recently left the company and founded Core Automation. Like Silver, he sees continuous learning as one of the final missing pieces before true AI emerges. Humans don't have a separate learning mode; everything happens at once, Tworek explained in a recent podcast interview. As long as models don't learn directly from data, their capabilities will remain limited. That kind of fundamental research, he added, is no longer possible at a heavily commercialized company like OpenAI.