A decisive truth, applied directly to the forehead of every AI company that spent billions scraping the internet: David Silver, the DeepMind legend who created AlphaZero, just raised \$1.1 billion for a startup that wants to build AI without any human training data whatsoever. The company is called Ineffable Intelligence, which is either a profoundly self-aware name or a warning label.

Sequoia Capital led the round. Google, Nvidia, the UK’s sovereign AI fund, and several other investors who presumably enjoy existential risk as a hobby all participated. Valuation: \$5.1 billion. For a company whose product is, essentially, “what if the AI just figured everything out by itself?”

🤚 The Open-Palm Premise

Silver’s thesis is elegantly terrifying: instead of training AI on terabytes of human-generated text — with all our biases, errors, and Yahoo Answers contributions — you let the AI learn entirely through reinforcement learning and self-play. No human data. No human examples. Just the machine, playing against itself, getting better.

If this sounds familiar, it’s because AlphaZero did exactly this with chess, Go, and shogi. It started knowing nothing but the rules, and within hours it was better than any human who has ever lived. The small, trivial difference is that Silver now wants to do this with… everything.

👐 The Two-Handed Paradigm Shift

Every major AI lab — OpenAI, Anthropic, Google — has built their frontier models on the same basic recipe: inhale the internet, digest it with transformers, fine-tune with human feedback. Silver is betting that recipe has a ceiling, and that the real breakthrough comes from AI that learns the way AlphaZero learned: from scratch, by pure exploration.

The implications are staggering. If it works, you don’t need the internet’s data. You don’t need human annotators. You don’t need to worry about copyright lawsuits from every newspaper and publisher on Earth. You just need compute, rules, and patience.

The “if it works” is doing Olympic-level heavy lifting in that paragraph.

👑 The Gold-Leaf Reckoning

Here’s the part that should keep you up at night: AlphaZero didn’t just match human chess — it developed strategies that humans had never conceived in 1,500 years of playing. It didn’t learn from us. It surpassed us by ignoring us entirely.

Now imagine that same principle applied to science, medicine, engineering, or warfare. An intelligence that doesn’t think like us because it was never trained on us. It’s the kind of idea that either wins the Nobel Prize or becomes the opening paragraph of a very concerning Wikipedia article titled “The Incident.”

\$1.1 billion says it’s the former. Probably.

Ineffable Intelligence could not be reached for comment, possibly because it was busy teaching itself to communicate without our help.