Primary source宝玉
View sourceEvent ·Multi-source
METR 评估:Anthropic 模型能力进步速度约为原来的 2 倍
METR 于 11 月 5 日发布报告,称 2025 年 10 月以来 Anthropic 模型的 AECI 测试分数增速是此前的 1.9 倍(95% 置信区间 1.5 到 2.8 倍)。METR 猜测加速的约八成来自 Anthropic 用 AI 做研发,但员工内部没有共识,且与训练算力增长无关。Anthropic 联合创始人 Sam McCandlish 回应称 METR 的许多结论与公司判断一致,未来几周会发布自己的报告,对照 RSP 安全政策说明进展。公开版报告中被 Anthropic 删减了大部分关键证据,公众无法自行核对。
Current read
METR 用测试数据算了下,发现 Anthropic 模型变强的速度快了一倍,主因可能就是拿 AI 来研发 AI。报告细节挺多,连内部员工自评效率提升 12 倍都写了。
11 sources82° AI heatLast updated 2026/11/5 22:12:36
Evidence
11 sourcesRelated source 1Thomas Wolf
View sourceRelated source 2METR
View sourceRelated source 3IT之家
View sourceRelated source 4Anthropic
View sourceRelated source 5rohanpaul_ai
View sourceRelated source 6Genk
View sourceRelated source 7kimmonismus
View sourceRelated source 8Andrew Curran
View sourceRelated source 9thsottiaux
View sourceRelated source 10techcrunch
View sourceConflict check
Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.