METR：Research（网页）·· 13 小时前AI 评分56
METR 提出支出视界指标，以 NanoGPT 竞速度量 AI 智能体优化能力
Expenditure Horizon: Measuring Optimization Ability, with an Application to NanoGPT July 21, 2026 We propose a measure of an AI agent’s optimization ability with an "expenditure horizon." We give an empirical illustration from the NanoGPT speedrun. Read more
AI 导读
METR 提出"支出视界"指标，用人类与智能体在优化问题上成本收益曲线的交点来量化智能体优化能力。以 NanoGPT speedrun 为例，人类每提升 1% 训练速度约需 16 小时劳动。
来源：METR：Research（网页） · metr.org