Dream-RSI

A University of Maryland-led framework (with UVA and Google DeepMind co-authors) for the exploration bottleneck in recursive self-improvement: fixed exploration strategies fail as search spaces grow, while optimising them online means searching a vast meta-space under delayed, expensive feedback. Dream-RSI makes exploration explicit and programmable through a lightweight orchestration layer and trains exploration policies inside replayable "dream" simulators of the task world rather than in the expensive world itself, leaving the underlying agent untouched. About 400 GitHub stars in its first week.