Ineffable Intelligence

Ineffable Intelligence researches reinforcement-learning systems that acquire knowledge through experience.

Research dossier · Reviewed 2026-09-21

Research and release status

The reviewed materials describe a research mission and infrastructure partnership, without a verified public model release.

Learning through interaction

David Silver founded Ineffable to pursue systems that learn from acting in an environment rather than primarily absorbing human-produced data. The lab calls its target a superlearner. Its claims about eventual generality and continual improvement describe ambitions, not established results. [1][2]

Infrastructure for reinforcement learning

In May 2026 the London-based lab announced engineering work with NVIDIA on training infrastructure. The announcement emphasizes the repeated cycle of acting, observing, scoring, and updating, and the resulting demands on memory, interconnects, and serving. [2]

What to watch

The relevant evidence will be learning results across environments, efficiency measurements, and demonstrations that knowledge transfers beyond the training setting. A hardware partnership supports the research effort but does not settle those questions. [2]

Sources

[1] Mission and founder's note ↗Ineffable Intelligence · accessed 2026-09-21
[2] NVIDIA infrastructure collaboration ↗Ineffable Intelligence and NVIDIA · accessed 2026-09-21