What does it mean to build a mind that is not a copy of yourself?
Today, we are pleased to announce that Sequoia has partnered with David Silver and Ineffable Intelligence. Ineffable Intelligence is a new AI research institute based in London with a singular mission: to make first contact with superintelligence.
Ineffable is building what David calls superlearners: systems that discover everything from elementary motor skills to deep intellectual breakthroughs directly through their own experiences. There is no pre-training. No imitation. They are just agents endlessly learning from the consequences of their own actions in a world built to teach them that.
Reinforcement learning-based superlearners have the potential to rediscover and surpass the greatest inventions in human history, including language, science, mathematics, and technology. Imagine a machine that derives the laws of physics from first principles. It invents new areas of mathematics that we have never thought of. It designs materials, medicines, and computers that we don’t yet have the vocabulary to describe. This is the prize David is trying to get.
another way
The current generation of AI was built by training across the human internet. That is an extraordinary achievement. However, systems trained on human data may also have fundamental limitations.
Ineffable Intelligence scales reinforcement learning from a clean base. No prior training or human data needed to shortcut the system. Guided by the age of experience as a north star, David proves that agents trained purely from the environment can develop inhuman strategies for reasoning about problems they don’t yet know how to solve.
David led some of the decisive breakthroughs in deep reinforcement learning. The most famous is Go, which is devilishly difficult. Go is the ultimate test of machine intelligence, as it cannot be brute-forced by computers. The combinatorially explosive O(10^170) valid disk positions far exceed the O(10^80) atoms in the observable universe. It was thought that it was simply too difficult to solve with machines.
At DeepMind, David drove a key breakthrough that ultimately solved the game of Go: self-play. Self-play led to an ~800 ELO point jump leading to the historic AlphaGo vs. Lee Sedol showdown in March 2016. David took this idea even further with AlphaGo Zero. By completely eliminating human pre-training and learning purely through self-play, the system’s ELO rating increased from ~3,700 to 5,000+. The result is a system that reaches decidedly superhuman performance and has somewhat inhuman mannerisms.
That’s the pedigree David has spent his career building. He was the lead researcher and technical force behind DeepMind’s Alpha series, and during its glory years his approach was the dominant paradigm, including AlphaGo, AlphaZero, AlphaStar, and AlphaProof.
When the LLM came along, David didn’t stop believing. He is one of the few people on the planet with the conviction, technical depth, and team to scale reinforcement learning.
what will happen next
The task ahead is difficult, the timeline for achieving superintelligence is uncertain, and the stakes are truly contrarian. That’s exactly what excites us. The biggest breakthroughs in AI always come from people willing to ignore consensus. David has defied more consensus than almost anyone in this field.
We are honored to co-lead the first round of Ineffable and partner with David on perhaps the most ambitious scientific mission of our generation.
