hillclimb
Training data for recursive self-improvement
Hillclimb is a virtual lab where AI systems can run experiments continuously and improve their research capabilities. It focuses on two problems: aggregating human research data and scaling the creation of reinforcement learning environments. The environments it produces are sold to frontier labs, and the company is hiring for roles in environments and post-training.
The homepage describes a mission centered on recursive self-improvement, with each new environment intended to expand the range of tasks available for training and evaluation. The company is backed by investors and individuals associated with major AI labs, and it directs inquiries about model research capabilities to its contact address.
12 alternatives to hillclimb
Ranked by how well each tool replaces hillclimb: shared features, audience, price and popularity.
Custom data for long-horizon reasoning, software engineering, and data science.
56 out of 100 match—Autoresearch as a service: turn a measurable target into thousands of directed experiments
53 out of 100 match—Building a new class of interpretable AI systems and foundation models that humans can
52 out of 100 match—