Concept系统级评估 / System-level Evaluation
Agentic system evaluation
Methods for evaluating a complete AI system composed of models, tools, and software, examining tool invocation, permission boundaries, error accumulation across multi-step tasks, and attack surfaces introduced by external dependencies.
Timeline (last 90 days)
Oct 19
Hugging Face's Irene Solaiman notes that the industry is shifting from simple model-level benchmarks to agentic system evaluations, but testing protocols lack consensus, making results hard to reproduce.
Unverified50%