New Paper: Towards a science of AI agent reliability

In three linesA new paper investigates AI agent reliability by quantifying the gap between claimed capabilities and actual performance. The study proposes methods to measure this divergence and improve the robustness of agent systems.

Summary generated by Claude — human-verified