David Silver’s $1.1B bet on an AI that doesn’t need us

David Silver, the guy who helped pioneer reinforcement learning at DeepMind and co-authored the AlphaGo paper, has just pulled off one of the biggest early-stage raises I’ve seen in a while. His new lab, Ineffable Intelligence, closed a $1.1 billion funding round at a $5.1 billion valuation. The company is only a few months old.

Let that sink in. A lab with no product, no revenue, and barely any public output is worth over five billion dollars. That’s not just hype — that’s a statement of intent from the investors backing it.

The core pitch is simple but ambitious: build an AI that learns entirely without human-generated data. No scraping Reddit, no labeling ImageNet images, no fine-tuning on Wikipedia dumps. The model figures things out through interaction and self-generated experience. This is basically the purest form of reinforcement learning — the kind Silver spent years researching at DeepMind, where agents learn by playing games millions of times without any human demonstration.

But scaling that from Go boards and Atari cartridges to real-world usefulness is a completely different challenge. The AlphaGo system needed massive compute and carefully designed reward functions. Ineffable Intelligence is essentially trying to do that for general intelligence, without humans holding its hand at any stage.

I’m skeptical, but not dismissive. The reinforcement learning approach has been tried before in various forms — from OpenAI’s early Dota bots to DeepMind’s own work on game-playing agents. None of them produced a general-purpose system. The leap from mastering StarCraft to understanding natural language or common sense reasoning is enormous.

Still, Silver has a track record that commands attention. He was instrumental in the breakthroughs that made DeepMind a household name. If anyone has a shot at making this work, it’s probably him. The question is whether pure self-play and reward optimization can scale to the messiness of the real world.

$1.1 billion is a lot of runway. Even with the insane cost of training large models, that kind of funding buys years of research time. The valuation is obviously inflated — there’s no way a pre-product company is worth $5B by any traditional metric — but in the current AI funding climate, it’s almost expected.

Ineffable Intelligence is still hiring and hasn’t published any technical details yet. The name itself is a bit pretentious, but I’ll let it slide given Silver’s credentials. What matters is whether they can actually deliver something that works outside a controlled environment.

For now, it’s a huge bet on a specific vision of AI progress — one that doesn’t rely on hoarding human data. That’s refreshing, honestly. The industry has become obsessed with data scale, and Silver is betting that intelligence can emerge from structure and interaction alone. We’ll see if he’s right.