David Silver walks away from DeepMind with $1.1B to build AI that doesn’t need humans

David Silver, one of the key minds behind DeepMind’s reinforcement learning breakthroughs, has pulled off one of the biggest early-stage raises I’ve seen in a while. His new lab, Ineffable Intelligence, just closed a $1.1 billion funding round at a $5.1 billion valuation.

That’s a lot of money for a company that, as far as anyone can tell, doesn’t have a product yet. The lab was founded only a few months ago. But Silver’s pitch is ambitious enough to attract that kind of capital: build an AI that learns without any human-generated data.

Most of today’s large language models and generative AI systems rely on massive datasets scraped from the internet — text, images, videos created by people. Even reinforcement learning from human feedback (RLHF), which Silver helped pioneer at DeepMind, still depends on human preferences to shape model behavior.

Silver wants to bypass all of that. The idea is to have the AI generate its own training data through interaction with simulated environments or self-play, then learn from those experiences without any human labeling or curation. If it works, you get systems that can scale indefinitely without bumping into the data bottleneck that every major lab is hitting right now.

This approach has been tried before in narrower domains. AlphaGo learned to play Go at superhuman level by playing millions of games against itself. But that was a closed system with a clear reward function — winning the game. Silver wants to generalize that to broader tasks where the reward signal isn’t obvious.

I’m skeptical about how far this can go without some form of human guidance. Pure self-supervised learning has limits. The world doesn’t come with a built-in scoreboard for most things. But if anyone has the credibility to push this frontier, it’s Silver. He was the lead researcher on AlphaGo and AlphaZero, and his work on reinforcement learning is foundational to the field.

$5.1 billion valuation for a pre-product company is eye-watering, no matter how you slice it. The AI hype cycle is still running hot, and investors are betting that Silver can replicate his DeepMind magic in a startup setting. Whether that works without DeepMind’s infrastructure and talent pool is another question.

The name “Ineffable Intelligence” is a bit much for my taste — sounds like something from a sci-fi novel about rogue AI. But maybe that’s intentional. Silver has always had a philosophical bent, often talking about intelligence as something that emerges rather than being programmed.

I’ll be watching to see if they actually ship something in the next 12 months. A billion dollars buys a lot of compute and talent, but it also comes with expectations. The AI research world is littered with well-funded labs that produced papers but no products. Silver’s track record suggests he can deliver, but the bar is set high.

If Ineffable Intelligence succeeds, it could fundamentally change how we think about training data and model scaling. If it fails, it’ll be one of the most expensive AI experiments in history. Either way, it’s going to be interesting.