Research lab that builds and sells training data for recursive self-improvement in AI by aggregating research data and automating RL environment creation.
Hillclimb aggregates human research data and automates the creation of reinforcement learning environments to produce training data that advances model capabilities toward recursive self-improvement. The company sells the training data it creates to frontier AI labs. It is backed by Tier 1 VCs and notable advisors including Jeff Dean and Paul Graham. hillclimb is a product of hillclimb.