understand.
research
and general intelligence
research
our story begins in academia, building on research that led to the founding of deepmind.
generative models have a truth problem, and iterating on the current technology will only make the lies more plausible.
we discovered an approach that solves this.
applied
and general intelligence
research
games use engines to simulate content (mechanics, gameplay, art, sound and narrative).
good content adheres to a particular "truth".
the artist's vision.
we can apply our technology to generate content that is consistent with a particular vision.
to build worlds
and general intelligence
and general intelligence
existing generative models can generate content on the fly. our technology generates consistent content.
we're working to scale our approach so we can generate entire game worlds, dynamically, in response to player actions.
a world that is internally consistent.
a world that is alive.
and general intelligence
and general intelligence
and general intelligence
we do this in service of a larger goal.
building an engine people use will help us validate our progress toward that goal.
the more compelling the world we generate,
the closer we are to modelling useful human intelligence.
the engine is just the beginning.
this is a project to build artificial general intelligence.