New layer lets AI testers decide early if one system beats another

Original: ParEvalLayer: When Partial LLM-Agent Evaluations Support a Decision

Source: arxiv.org ↗