rotative
mit tech review · 2026年11月20日

METR更新AI图表,揭示模型弱点

metr的更新报告显示ai模型快速改进,但仍在判断任务中挣扎。这影响了ai在研发领域的自动化能力。

read in English · 繁體中文

metr's updated report finds that ai models are still improving rapidly, with some capabilities continuing to improve along an exponential trajectory. however, the report also indicates that models struggle with tasks that require judgment, such as designing new ai evaluations and scoring philosophical arguments. metr evaluated models by having them attempt a large suite of tasks, each of which had also been completed by humans, and rated models according to the length of the tasks that they could complete. the updated plot adds more tasks, including some that are estimated to take humans far longer than the tasks in the original suite, and others that are more open-ended. this has significant implications for ai's ability to automate research and development, a key area of interest in the ai community.

read at source ↗

← rotative