New framework tests AI coding agents like scientific experiments

Original: An Experimental Design Approach to Evaluating Agentic AI's Autonomous Model Discovery

Source: arxiv.org ↗