David Silver left DeepMind and raised $1.1B to build an AI that doesn’t need us

David Silver, the guy who helped write the playbook for reinforcement learning at DeepMind, just raised $1.1 billion for a new lab called Ineffable Intelligence. The valuation sits at $5.1 billion. For a company that was founded a few months ago and hasn’t shipped a product, that’s a lot of trust—or a lot of hype.

Silver’s pitch is straightforward and ambitious: build an AI that learns without needing human-generated data. No scraping Reddit threads, no feeding it Wikipedia dumps, no hand-labeled image datasets. The idea is that true intelligence should emerge from interaction with the world, not from memorizing what humans already know.

This isn’t a new dream. DeepMind itself was built on reinforcement learning, where agents learn by trial and error in simulated environments. AlphaGo learned by playing millions of games against itself. The difference here is scale and scope—Silver wants to push that paradigm into domains where human data is sparse, noisy, or just doesn’t exist.

I find the name “Ineffable Intelligence” a bit much, but I get the vibe. They’re going for the idea that some forms of intelligence can’t be easily described or captured in static datasets. It’s a bet on emergence over curation.

The $1.1B round is massive, especially for a pre-product lab. For context, that’s more than most Series B rounds in AI right now. The $5.1B valuation implies investors are betting Silver can replicate his DeepMind track record—where he was instrumental in AlphaGo, AlphaZero, and the early reinforcement learning breakthroughs that eventually led to systems like AlphaFold.

But here’s the thing: reinforcement learning at scale is brutally hard. It works in games and simulators because you can run millions of episodes cheaply. In the real world, you don’t get do-overs. Robots break, physics is messy, and the cost of exploration is high. Silver’s team will need to crack sample efficiency in a way that hasn’t been done before.

There’s also the question of data quality. Human data is biased, but it’s also structured. An AI that learns entirely from scratch might discover solutions that are optimal but incomprehensible—or outright dangerous. The alignment problem doesn’t go away just because you stop using human labels.

Still, I respect the ambition. The AI field has become obsessed with scaling laws and bigger datasets, and Silver is basically saying “let’s try the hard way instead.” If anyone can pull it off, he’s on the short list. DeepMind’s early work showed that agents could discover strategies humans never thought of—like AlphaGo’s move 37 in game 2 against Lee Sedol.

The timing is interesting too. We’re seeing a split in AI research: one camp doubling down on large language models and massive supervised training, another pushing toward agents that learn from experience. Silver’s lab is firmly in the second camp, and this funding gives them runway to go deep.

Ineffable Intelligence hasn’t published a paper or demo yet, so we’re flying blind on technical details. The team is reportedly small, mostly ex-DeepMind researchers Silver handpicked. That’s both reassuring and worrying—small teams move fast, but also have less redundancy when things go wrong.

For now, the industry is watching. If Silver’s approach works, it could redefine how we think about AI training. If it doesn’t, it’ll be one of the most expensive experiments in AI history. Either way, it’s a bet worth taking.