NextFin News - In a move that signals a profound shift in the global race for artificial general intelligence, David Silver, the legendary researcher behind AlphaGo and AlphaZero, has officially departed Google DeepMind to launch a new venture, Ineffable Intelligence. According to The Decoder, Silver’s departure marks the end of a foundational era at the London-based AI lab, where he was one of the earliest employees and a primary architect of its most celebrated breakthroughs. The new startup, registered in London as of November 2025, is currently in an aggressive recruitment phase, seeking top-tier AI researchers and venture capital to pursue a vision of superintelligence that moves beyond the current industry obsession with Large Language Models (LLMs).
The timing of Silver’s exit is particularly significant as U.S. President Trump’s administration continues to emphasize American leadership in AI through deregulatory frameworks and accelerated infrastructure investment. While Google DeepMind remains a crown jewel of the Alphabet portfolio, the departure of its most prominent reinforcement learning expert suggests an internal divergence regarding the technical roadmap to superintelligence. Silver’s new firm, Ineffable Intelligence, is predicated on the belief that LLMs—which rely on predicting the next token based on existing human knowledge—are fundamentally incapable of achieving true superintelligence. Instead, Silver is doubling down on the "Era of Experience," a paradigm where AI agents learn through trial and error within simulated world models, effectively discovering knowledge that humans do not yet possess.
This strategic pivot reflects a broader trend of "architectural skepticism" among the pioneers of the field. Silver joins a growing list of elite defectors from major labs, including Ilya Sutskever, who left OpenAI to found Safe Superintelligence, and Jerry Tworek, another key OpenAI researcher who recently exited citing the limitations of static training models. The core of the argument, as articulated by Silver and his long-time collaborator Richard Sutton in their April 2025 research, is that current AI is "frozen" after training. In contrast, Ineffable Intelligence aims to build systems that exhibit continuous learning—adapting to environments over months or years without the need for massive, human-labeled datasets. This approach mirrors biological intelligence more closely than the current Transformer-based systems that dominate the market.
From a financial and industrial perspective, Silver’s move highlights the increasing difficulty of conducting fundamental, high-risk research within heavily commercialized entities. As Google and OpenAI face immense pressure to deliver quarterly returns on their multi-billion dollar GPU investments, their focus has naturally shifted toward refining LLMs for enterprise applications. However, the marginal utility of adding more parameters to LLMs appears to be diminishing. Data from recent industry benchmarks suggests that while reasoning capabilities are improving, the underlying architecture still struggles with causal reasoning and genuine novelty. By moving to a startup environment, Silver gains the agility to explore reinforcement learning at scale—a method that famously allowed AlphaZero to master chess in hours by playing against itself, rather than studying human games.
The geopolitical implications are equally stark. With Ineffable Intelligence based in London, the United Kingdom maintains a critical foothold in the next generation of AI development, even as U.S. President Trump’s policies aim to centralize AI compute power within the United States. The success of Silver’s venture could prove that the path to superintelligence is not merely a matter of who has the most H100 GPUs, but who possesses the most efficient learning algorithms. If Silver can replicate the self-learning success of AlphaGo on a general-purpose scale, the reliance on massive, energy-intensive data centers for training might be superseded by more efficient, autonomous learning agents.
Looking ahead, the industry should expect a "de-coupling" of AI development. On one side will be the "Scaling Hypothesis" camp, led by the major tech incumbents, who believe that more data and more compute will eventually bridge the gap to AGI. On the other side will be the "Algorithmic Innovation" camp, led by figures like Silver and Sutskever, who argue that a fundamental change in how machines learn is required. For investors, this represents a high-stakes gamble: the LLM-centric approach has already proven its commercial viability, but the reinforcement learning approach championed by Silver is the only one that has historically demonstrated the ability to surpass human-level performance in complex, closed-system environments. As 2026 progresses, the progress of Ineffable Intelligence will serve as a litmus test for whether the next leap in AI will be built on the words of the past or the experiences of the future.
Explore more exclusive insights at nextfin.ai.
