论文
42
AI 评分(两次独立评分的平均)METR 扩展时间跨度评测任务集
原标题:To better understand AI capabilities, we extended our time horizon suite with additional cheaply-verifiable tasks, and added tasks that a...
为了更好地理解 AI 能力,我们扩展了时间跨度评测套件,加入了更多可低成本验证的任务,并新增了难以验证或需要判断的任务。 我们在这些新增和现有任务上测试了公开模型,最高至 Opus 5.5 和 GPT-6 Astra。
中文译文尚未完成,先为你展示原文。
To better understand AI capabilities, we extended our time horizon suite with additional cheaply-verifiable tasks, and added tasks that are hard to verify or require judgment.
We tested public models up to Opus 5.5 and GPT-6 Astra on these new and existing tasks.
