Create

Sign in to ReadmeX

or

How we handle your data: Privacy Policy

ReadmeX

Your AI reading companion

What would you like to catch up on?

Pick a briefing, or ask your own question.

Your community and people briefings stay personal to you.

Story

METR says Anthropic's internal AI progress rate is nearly twice as fast

AI summary

METR estimated that progress on clearly defined tasks at Anthropic reached about 1.9 times its previous rate, with a 95% confidence interval of 1.5 to 2.8. METR said this means roughly two years of capability progress in one year—not doubled model capability or uniformly doubled employee productivity—and that AI-assisted research remains primarily human-assisted. Anthropic separately disclosed that it restricted certain capabilities in Claude Fable 5.2 and Mythos 5.2 that could directly accelerate AI development, while warning that its restrictions’ effectiveness has not been independently established.

In-depth insight

The assessment offers an independently measured signal of faster AI-assisted research at Anthropic, but an incomplete compute-allocation cross-check and Anthropic’s concern over commercial sensitivity limit what can be verified.

DoNews 11月10日消息 Anthropic公开承认已经采取措施限制了某些可能直接加速AI开发的能力。目前该公司正陷入一场相关的风险评估风波。

11月2日,Anthropic在招股书修订稿S-1/A中提到,根据内部估计及对未来模型的预期,公司可能很快达到《负责任扩展政策》定义的“由AI带来的显著加速进展”门槛。11月5日,非营利性人工智能安全研究机构METR发布的一份评估与Anthropic产生了争议。

METR表示,评估证据来自该机构研究人员在8月12日至10月29日解除到的Anthropic内部数据及相关员工。一些重要结论:在定义明确的任务上,模型能力进步速度可能已达到此前的约两倍。

AECI能力指标(Anthropic Epoch Capability Index,简称为Anthropic综合能力指数)的增长速度估计为此前的1.9倍,95%置信区间为1.5—2.8倍。这里的“两倍”指大致用一年完成过去两年的能力进步,不是模型能力翻倍,也不是员工生产率统一翻倍。

#Anthropic#METR#Claude Fable 5.2#Claude Mythos 5.2#Claude Opus 5.5#Anthropic Epoch Capability Index#Responsible Scaling Policy

Original · 3 min readRead the original
Group chat

No comments yet. Start the conversation.