Ineffable Intelligence: 1,1 mld dolarów na AI bez danych ludzkich

David Silver, the researcher behind AlphaGo, AlphaZero and MuZero, has left Google DeepMind and announced a new startup, Ineffable Intelligence. The company has raised $1.1 billion across seed and Series A rounds — one of the largest first funding rounds in AI history.

Ineffable Intelligence sets itself apart from typical large language model labs. Silver wants to build systems that learn primarily from their own interaction with an environment, rather than from massive scrapes of internet text and images.

Key facts

• Funding: ~$1.1 billion (seed + Series A); lead investors not publicly disclosed in the launch communication.

• Founder: David Silver, former lead of reinforcement learning at Google DeepMind, co-author of AlphaGo, AlphaZero and MuZero.

• Technical thesis: build general-purpose models through environment interaction (the "era of experience") rather than scaling pretraining data.

• Status: hiring phase; no public models or demonstrations yet.

• Domain: ineffable.ai — currently limited to a logo and short manifesto.

Who is David Silver

For more than a decade Silver led reinforcement learning research at DeepMind. His team built AlphaGo, the system that defeated Lee Sedol in Go in 2016, and AlphaZero, which learns games from scratch with no expert knowledge. He later worked on MuZero, where the agent learns a model of its environment without being told the rules.

In 2024 Silver and Richard Sutton, one of the founders of reinforcement learning, published the essay "Welcome to the Era of Experience". The text became a manifesto for the view that the next step in AI is not more pretraining on human data but agents that primarily learn through direct contact with the world.

What Ineffable Intelligence aims to build

Public information suggests Ineffable Intelligence wants to build a general AI system in the "era of experience" paradigm. In practice this means an architecture in which the agent gathers its own data through interaction with an environment (digital, physical or simulated), plans, decides and corrects its strategies.

The company has not disclosed whether it intends to work with physical robots, software agents, large-scale simulations or a combination of all three. The name "Ineffable" — incapable of being expressed in words — points back to Silver's thesis that important forms of intelligence go beyond what can be captured in text corpora.

Why $1.1B at this stage is unusual

Over the past two years a handful of AI startups have raised more than $1B at a very early stage — including Ilya Sutskever's Safe Superintelligence and Mira Murati's Thinking Machines. Ineffable Intelligence joins that short list but stands out for its sharply defined research thesis: instead of broad statements about "safe superintelligence", it offers a concrete hypothesis about what the next generation of models should look like.

The round size also reflects physical costs: training agents that learn from experience requires dense compute infrastructure (GPU/TPU clusters) and, most likely, large-scale simulation environments. Without capital of this magnitude, AlphaZero-class experiments would be hard to reproduce outside top labs.

Why it matters

Ineffable Intelligence is a real-world test of a hypothesis that has been growing inside the reinforcement learning community for years but was overshadowed by the success of language models: that the limit of scaling human internet data is near, and the next jump in AI capability requires learning from experience. If Silver pulls this off operationally, it will reshape robotics, software agents and decision systems. If he does not, it will be the best-funded confirmation that the hybrid of large language models with reinforcement learning remains the dominant paradigm for the next several years. From a market perspective, a round of this size at this stage tightens competition for reinforcement learning talent — and DeepMind loses one of its most recognizable researchers.