David Silver needs no introduction. The UCL professor built AlphaGo at DeepMind — the system that beat the world’s best Go player in 2016. Now he has a new venture, and it’s ambitious: Ineffable Intelligence wants to build AI systems that don’t learn from human data, but from their own experience.
What Makes Ineffable Intelligence Different
Most large language models — Claude, GPT, Gemini — are trained on massive amounts of human text. That works well, but Silver argues it’s only the easier half of the problem. The harder half is building systems that discover new knowledge on their own.
Ineffable is betting on reinforcement learning at scale. Instead of reading text, their models learn through trial and error — similar to AlphaGo, but thinking much bigger.
Nvidia Goes All In — With Hardware and Engineers
On May 13, Nvidia announced a strategic partnership with Ineffable. This isn’t just about money: engineers from both companies are collaborating on infrastructure for reinforcement learning at scale. They’ll be using Nvidia’s Grace Blackwell chips and the upcoming Vera Rubin platform.
Jensen Huang put it simply: ‘The next frontier of AI is superlearners — systems that learn continuously from experience.‘
A Record $1.1 Billion Seed Round
The numbers are staggering: Ineffable closed a $1.1 billion seed round — a record. Led by Sequoia and Lightspeed, with participation from Nvidia, Google, DST Global, and the UK’s Sovereign AI Fund. The startup was only founded in late 2025.
What This Means
This deal is a signal. Nvidia, the most important hardware supplier in AI, isn’t just betting on the next generation of chips — they’re also backing a fundamentally different approach to training. If Silver is right that current LLMs are hitting a ceiling, reinforcement learning could be the next big leap.
Whether this actually leads to ‘superintelligence,’ as Ineffable aspires to, is another question entirely. But investors are clearly betting it will.
Sources: CNBC