David Silver, a researcher associated with DeepMind and known for work on AlphaZero-style reinforcement learning, has raised $1.1 billion for a London-based lab called Ineffable Intelligence. The company was valued at $5.1 billion after only months of operation.
What the lab intends to do
Ineffable Intelligence focuses on developing large-scale models that learn predominantly through self-play, simulation and experience rather than training on large corpora of human-generated text, labeled datasets or demonstrations. The approach echoes the AlphaZero playbook: discover skills through the model’s own trial-and-error instead of copying human examples.
Why this matters
Today's widely used large language models (LLMs) and many other AI systems rely heavily on human-origin internet content — books, code, posts, labels, demonstrations and synthetic imitations of human behavior. Silver and his team argue there is an effective ceiling to what such human-derived data can deliver, and that the next phase of scaling intelligence may rest more on compute, simulation and relentless empirical optimization.
Relation to past successes
Silver’s prior work is connected to AlphaZero and AlphaGo, where self-play enabled strong performance without storing human game records. Investors are betting that the same underlying logic might generalize from mastering games to discovering broader forms of intelligence.
Potential implications
If the lab's approach proves effective, it could shift emphasis in model development away from massive human-data collection and annotation toward computation-heavy training regimes, realistic simulations and self-improving algorithms. Practical challenges remain, including cost, engineering obstacles and questions about real-world applicability.
Conclusion
The $1.1 billion raise and $5.1 billion valuation signal investor confidence in exploring routes to AI that do not primarily depend on human-generated training data. Ineffable Intelligence’s work will be an important test case in the debate over whether future AI should emulate human data or discover intelligence via autonomous experience.
