Karotte goes open source for building reinforcement-learning environments

Karotte’s developers announced they are open-sourcing the framework for building reinforcement-learning environments. They say it has been used for a year to create machine-learning-engineering tasks for frontier labs and hardened through more than 1 million evaluation runs and red-teaming. A team member also attributed $15 million in recent revenue to environments built with Karotte, a figure that has not been independently verified.