Industry

预测:多条S曲线叠加为何能产出能力与GDP的直线增长

Why we picked it

这是条一分钟的推文,但讲清楚了 scaling 争论里的关键机制:单条曲线会饱和,多条 S 曲线加新指数叠加起来就成了直线,GDP 也是这么长出来的。想跟人聊 AI 能力外推时,这条提供了一个现成的框架,比背数字更好用。

一条来自 X 用户 tszzl 的推文观察到一个反直觉现象:单个能力的增长往往呈 S 型(sigmoid)并逐渐进入边际收益递减,但把多条 sigmoid 曲线与零星出现的新指数曲线叠加后,整体能力曲线(capabilities graphs)却呈现近似直线的形态。作者指出 GDP 增长曲线也存在同样的结构。这一观察把"指数增长预期"与"递减收益现实"之间的张力归因于曲线叠加效应:一条曲线的放缓被另一条新曲线的起点补上。推文未引用具体基准名称或数字,属于对缩放定律(scaling laws)与长期增长叙事的定性讨论。

Body

这类观点常见于关于 AI 能力外推的争论:怀疑方以单条曲线的饱和为依据预测平台期,而乐观方以新范式的接力来解释持续直线外推。作者以 GDP 曲线作类比,把宏观经济的长期增长也理解为技术与新范式逐波接续的结果。该推文本身没有提出新数据或新模型,更适合作为理解增长曲线争论的视角参考。

Original · roon

it really is pretty strange how a bunch of sigmoids tapering out on diminishing return curves combine with new exponentials here and there all to produce straight lines on capabilities graphs. ditto for the gdp curves

Industry Coordinates & Source Verification

This report is categorized under AITOP's "Industry" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include scaling laws, reasoning models.

Source attribution: curated from roon (@tszzl).