Sequoia is co-leading the first funding round for Ineffable Intelligence, a London AI lab founded by David Silver with one stated goal: build superintelligence from scratch using pure reinforcement learning. No pre-training. No human data. The system, which Silver calls a superlearner, learns entirely from consequences of its own actions inside a purpose-built environment.
Silver's credentials make this worth taking seriously. He led the AlphaGo team at DeepMind, drove the self-play breakthrough that produced an 800 ELO-point leap before the March 2016 match against Lee Sedol, then stripped out human pre-training entirely with AlphaGo Zero, pushing the system's ELO from roughly 3,700 to above 5,000. Go has approximately 10^170 legal board positions, more than the 10^80 atoms in the observable universe. Machines were not supposed to solve it. Silver's argument, supported by that track record, is that removing human imitation produces non-human strategies, and that the same principle scales beyond games.
The full piece is worth reading for Silver's specific technical framing of what a superlearner could rediscover: language, mathematics, physics from first principles, materials and medicines that current vocabulary cannot yet describe. Sequoia's Sonya Huang and Alfred Lin call the bet genuinely contrarian. The timeline to superintelligence is unspecified. What the original lays out is the precise architectural logic for why starting from zero, not from the internet, might be the only path to getting there.
[READ ORIGINAL →]