1 paper
Bo Deng, Kang Zhou, Lifan Guo +6
Most agent benchmarks evaluate tasks independently and cannot measure whether experience from one task helps with later tasks. Existing self-evolution benchmarks do not jointly cov…