Gary Marcus 紧急澄清：别慌，METR 图被过度反应了

Gary Marcus 发推紧急叫停对 METR 新图的恐慌式解读。他指出几个被忽略的关键背景：①Claude Code 是真实进展，Mythos 很可能建立在其基础上；②图表测量的是达到 50% 成功率，不是 90% 或 100%；③仔细读图会发现很多方法其实还达不到 50%。Marcus 以冷静的数据分析视角平衡了 AI 圈常见的过度炒作和过度恐慌。这条推文本质上是科学素养教育：不要根据二手解读做判断，要回到原始数据。

Gary Marcus: 不要对 Mythos/METR 图表恐慌

原文 / Original

PLEASE DO NOT PANIC about the Mythos/METR graph that everyone is panicking about.

Progress is being made but people are totally overreacting.

Here's some context that is being left out from nearly every comment on that graph.

[Replies from the thread include context about how the graph's methodology affects interpretation, with security researchers sifting through AI slop findings, and that the 50% reliability catch is real but the slope (doubling-every-N-months) is what drives the alarm. Some note that the graph doesn't consider cost factors.]

Gary Marcus: 不要对 Mythos/METR 图表恐慌

原文 / Original

继续阅读