AlphaGo's Creator Thinks AI Is Going the Wrong Way

David Silver, the researcher who built AlphaGo, thinks the AI industry has taken a wrong turn — and he's putting roughly a billion dollars behind that opinion. His new company is focused on building what he calls "superlearners": AI systems that genuinely learn from experience rather than pattern-matching on static training data. When the guy who cracked Go says everyone else is doing it wrong, that's worth paying attention to.

The Man Who Made AI History in 2016

Silver is the lead architect of AlphaGo, the DeepMind system that defeated Lee Sedol — then the world's best Go player — back in 2016. That win wasn't a PR stunt. It was a genuine inflection point that proved AI could master domains long considered off-limits to machines. Silver has spent his career working on deep reinforcement learning, a fundamentally different approach from the large language models that now dominate every AI conversation.

A Billion-Dollar Bet Against the LLM Consensus

Silver has founded a new startup backed by approximately one billion dollars in funding, according to Wired. The company's core goal is to develop AI agents powered by deep reinforcement learning — systems that improve through direct experience in an environment, not just by ingesting massive text datasets. The startup doesn't have a public name or product yet, but the financial backing alone puts it on the industry map immediately. Silver's central argument is that the current LLM paradigm — training enormous models on internet text — has hard fundamental limits that more compute and more data simply cannot overcome.

Saying Out Loud What Many Researchers Whisper Privately

Silver's critique of the LLM-first approach isn't new in research circles, but few people with his track record are saying it publicly with this level of financial commitment behind them. His thesis is pointed: the path to artificial general intelligence doesn't run through bigger language models — it runs through systems that can explore, adapt, and improve autonomously. If he's right, that's a real problem for OpenAI, Anthropic, and Google's current research directions. If he's wrong, it's a very expensive disagreement.

What This Means for the Broader AI Industry

Having someone with Silver's credibility move serious capital toward reinforcement learning could meaningfully shift research priorities at a moment when almost every major lab is doubling down on transformers. The ripple effects could include:

• Renewed institutional interest in non-LLM approaches to AI development
• Pressure on major labs to diversify their research bets beyond scaling
• A sharper public debate about whether parameter scaling actually has a ceiling

The industry has been running a "bigger is better" playbook for two years straight. Silver is explicitly betting that playbook has already peaked.

If the man who taught a machine to master Go is right, the next real breakthrough in AI won't come from a larger language model — it'll come from a system that actually knows how to learn.

Source: Wired