Signal觉醒信号Oct 29, 20262026年10月29日Note短评

Copying the frontier is cheap; building the next is the hard part.复制前沿很便宜，造出下一个前沿才是最难的一步。

A bright, sunlit gallery. On a glowing pedestal stands an original, intricate crystalline sculpture. Beside it a faceless figure, seen from behind, proudly holds up a flat, too-perfect flattened copy of that sculpture, with the brand mark stamped small in its corner. In the soft shadow behind them an empty easel holds a blank white canvas — the next original no one has made yet.明亮阳光下的展厅：发光的基座上立着一件独一无二、繁复的原创水晶雕塑；旁边一个背对镜头、看不清面孔的人影，骄傲地举起一块把那件雕塑压平复制出来的面板——过分完美，一看就是复制品，角上小小地印着的标志。人影身后的阴影里，一个空画架上摆着一张全白的空白画布——那个还没人做出来的『下一个原创』。Ref.: RoadMaster19 / Wikimedia Commons · CC BY 4.0Ref.: RoadMaster19 / Wikimedia Commons · CC BY 4.0

DeepSeek and Kimi just matched the frontier cheaply, and The Information reports they did it by distilling OpenAI's Astra. On Oct 29 it called Astra the top distillation target for China's AI labs: DeepSeek's V4.1-Flash, out Oct 17, and Moonshot's Kimi K3.1 Pro, out Oct 23, reportedly trained on Astra's outputs; Kimi K3.1 Pro ranks 4th on Artificial Analysis' intelligence index. The "caught up at one-tenth the cost" story spread instantly.

Strip it down. Distillation trains a new model on a stronger one's outputs: you buy its behavior, not the R&D that produced it. A leaderboard rank measures how close you are to the teacher, not whether you can make the next capability no one has built yet. The side that paid to blaze the frontier is the one being distilled, and originating it has no cheap shortcut. Second-order: once anyone can distill the leader, the leader shuts the door — OpenAI is already tightening rules against distillation, and the US CISA has issued an advisory. Capability caught up this way is borrowed and revocable.

The strongest counter: DeepSeek and Moonshot have made real original contributions in training efficiency, MoE architecture and open weights, and genuinely lead on cost and open deployment. So the judgment narrows to one point: cheaply matching the frontier is no proof of independent frontier R&D capability.

This breaks if a Chinese lab is first to build a genuinely new capability with nothing to distill from, or keeps nearing the frontier after its distillation access is cut off. Until then, separate "matched today's frontier" from "can define tomorrow's" — and know which one you're relying on.

DeepSeek、Kimi 便宜追平了最前沿的模型，The Information 报道称他们是蒸馏了 OpenAI 的 Astra：10 月 29 日它把 Astra 称作中国 AI 实验室的头号『蒸馏目标』，DeepSeek 的 V4.1-Flash（10 月 17 日发布）和 Moonshot 的 Kimi K3.1 Pro（10 月 23 日发布）据称都用了 Astra 的输出来训练；Kimi K3.1 Pro 在 Artificial Analysis 智能榜上排到第 4。『十分之一成本就追平』的说法立刻刷屏。

拆到底：蒸馏是拿更强模型的输出训练新模型，你买到的是它的『行为』，不是产出这种行为的研发能力。榜单衡量的是『离老师有多近』，不是『能不能做出还没人做出来的下一个能力』。花大钱蹚出前沿的是被蒸馏的一方，而原创前沿没有便宜的捷径。二阶效应随之而来：既然谁都能蒸馏领先者，领先者就会关门——OpenAI 已在收紧规则封堵蒸馏，连美国 CISA 都发了预警。靠蒸馏追来的能力，既是借来的，也随时可能被切断。

最强的反方：DeepSeek 和 Moonshot 在训练效率、MoE 架构和开源权重上有实打实的原创贡献，论成本和开放部署确实领先。所以判断只收窄到一点：便宜追平前沿，证明不了独立的前沿研发能力。

如果某家中国实验室在无从蒸馏时率先做出一个全新能力，或在被切断蒸馏通道后仍能逼近前沿，这条判断就需要修正。在那之前，看到『又便宜又追平』，先分清『追上今天的前沿』和『能定义明天的前沿』——你要依赖哪一个，心里得有数。

Built on the WAKENEX concept:所用的 WAKENEX 概念： Results alone don't prove capability能力不能由结果单独证明

WAKENEX / NOTESWAKENEX / 短评

Add WAKENEX as a preferred source on Google在 Google 把 WAKENEX 设为首选来源

READ NEXT接着读

Google's Multi-Party Compute buys stronger key security at one new dependence.谷歌的 Multi-Party Compute 用更强的密钥安全，换来一项新的依赖。

Most-used on OpenRouter is a price signal, not proof of capability.在 OpenRouter 上用量第一，是价格信号，不是能力的证明。

The Fed's 10-2 hold isn't a verdict that inflation is beaten.美联储 10 比 2 的按兵不动，并不是通胀已被制服的判决。