他推动 AI 基准测试成为科学而非“暗术”
摘要
有研究者主张把 AI 基准测试从“暗术”变成一门科学，以规范衡量模型能力的标准测试。文章以 1987 年美国国防高级研究计划局（DARPA）首个自动语音转文本基准为背景，讨论基准既是测量工具、也会主导整个领域行为的两面性。
摘要由模型根据原文撰写，以原文为准。
正文 · 原文
Benchmarks are the rulers of AI, in both senses of the word. As measuring sticks, they work according to an extremely simple logic: If you want to know what an AI model can do, devise a standardized test to gauge it. But AI benchmarks can also dominate the behavior of the entire field. In 1987, the Defense Advanced Research Projects Agency’s first benchmark on automatic speech-to-text…
注意转载授权 · 阅读原文
来源：Quanta Magazine · quantamagazine.org
引用原文
John Pavlus. He Wants AI Benchmarking To Be a Science, Not a Dark Art[EB/OL]. (2026-10-16)[访问日期]. https://www.quantamagazine.org/he-wants-ai-benchmarking-to-be-a-science-not-a-dark-art-20261016/.