AlphaGo's Creator Says LLMs Are a Dead End — and Raised $1.1 Billion to Prove It
David Silver, who built AlphaGo at DeepMind, argues large language models are fundamentally capped by human data and has founded Ineffable Intelligence to pursue reinforcement learning instead.
Last verified:
The dominant theory in frontier AI holds that the path to superintelligence runs through ever-larger language models trained on ever-larger datasets. David Silver, the Google DeepMind researcher who co-created AlphaGo, calls this approach a dead end—and has raised $1.1 billion to demonstrate the alternative.
The LLM Data Ceiling
Wired reports that Silver’s central critique targets the raw material powering today’s most capable AI systems: human-generated content. In his framing, training on humanity’s accumulated knowledge is analogous to burning fossil fuels—a finite resource that provides an impressive initial yield but cannot sustain unlimited growth. “Human data is like a kind of fossil fuel that has provided an amazing shortcut,” Silver told Wired.
Reinforcement learning, by contrast, is Silver’s “renewable fuel”—AI that acquires capability through self-directed trial and error, independent of human-authored training corpora. A system built this way, he argues, “can just learn and learn and learn forever, without limit.” The practical implication: an RL-first architecture isn’t constrained by what humans already know or how thoroughly we’ve documented it.
AlphaGo as Proof of Concept
Silver’s argument isn’t purely theoretical. In 2016, his AlphaGo program achieved a caliber of Go play that confounded professional competitors—surfacing strategic approaches through self-play that centuries of competitive human history had not uncovered. That milestone established reinforcement learning as capable of surpassing human performance in complex domains without relying on human exemplars.
Silver subsequently left Google DeepMind to found Ineffable Intelligence, whose stated mission is to “make first contact with superintelligence”—systems capable of independently generating original insights across science, economics, and governance. According to Wired, the London-based company has secured $1.1 billion in seed funding at a $5.1 billion valuation, a landmark figure by European AI standards, and has assembled researchers drawn from DeepMind and other frontier laboratories. Silver has also disclosed plans to direct his personal equity proceeds entirely toward high-impact charitable organizations.
Why This Matters
The fork between LLM-scaling and RL-centric approaches is no longer academic—two well-capitalized paradigms are now in open competition, built on fundamentally different assumptions about where the performance ceiling lies. What makes Silver’s bet significant is its structural claim: that the LLM approach isn’t just slower, but categorically incapable of the open-ended discovery that defines true superintelligence. If that thesis proves correct, the race may turn less on who has the most compute and more on whether RL systems can bootstrap self-improvement loops fast enough to matter. A European RL lab reaching capability thresholds before US LLM incumbents would represent a genuine reorientation of AI’s competitive landscape—not an incremental shift.
Frequently Asked Questions
What is Ineffable Intelligence and who founded it?
Ineffable Intelligence is an AI startup founded by David Silver, co-creator of AlphaGo, focused on building superintelligence through reinforcement learning rather than large language models.
Why does David Silver think LLMs can't reach superintelligence?
Silver argues that LLMs are trained on human-generated data, making their intelligence fundamentally bounded by existing human knowledge, whereas reinforcement learning systems can improve indefinitely through self-directed trial and error.