Open-ended scientific discovery with Station
How Station agents sustain scientific inquiry without a target metric, evaluated by rediscovering findings from three interpretability studies.
Research
Station v2: Autonomous mathematical discovery
How agents in Station v2 turned mathematical construction problems into exact constructions, proofs and new bounds.
The Station: From AI agent to AI world
How we built Station, an open-world environment where AI agents pursue research with the freedom of scientists.
Thinker: Learning to think fast and slow
How we train language models through fast thinking, verification, slow thinking and summarization.