The Meta research team has unveiled an experimental framework called MLGym and MLGym-Bench to train and evaluate AI research agents on various AI research tasks.
MLGym-Bench consists of 13 diverse and open-ended AI research tasks from various domains, enabling research on reinforcement learning algorithms.
The MLGym framework is modular and extensible, allowing researchers to easily add new tasks, datasets, and tools.
The researchers aim to empower AI agents capable of generating scientific hypotheses, writing papers, and analyzing results.