Understanding is an open problem.
We’re pursuing the foundations of general intelligence: systems that understand the world, reason about unfamiliar problems, and learn to act within it. Our long-term ambition is superintelligence in service of humanity.
Publications & technical reportsToward deeper intelligence.
World understanding
Representations that connect perception, context, cause and consequence.
Reasoning & planning
General reasoning, long-horizon planning and learning to solve unfamiliar problems.
Agents & autonomous systems
Systems that maintain context, use tools and pursue goals with meaningful human oversight.
Architectures & AI systems
Model architectures, learning efficiency and reliable infrastructure for advanced AI.
Safety & human agency
Alignment, evaluation and human control as models and agents become more capable.
The work, in detail.
Publication status
Our first reports are still ahead.
We haven’t published research papers or technical reports yet. The focus pages describe our questions and direction; they do not claim completed scientific results.
Build. Examine.
Learn.
Useful research should make its assumptions, methods and limitations understandable. We intend to share technical work with the context needed to evaluate it.
Safety & evaluation