AIDE^2: Recursive Self-Improvement for AI Research Agents

AIDE^2 implements a recursive loop where a frontier AI research agent proposes, benchmarks, and accepts rewrites of its own code. This system aims to counter diminishing R&D returns by allowing agents to optimize their own research efficiency.