今日问题
Daily Question
当 AI 让越来越多能力变得 trainable，什么东西仍然长期有价值？
When AI makes more abilities trainable, what remains valuable over the long run?
30 秒答案
30-Second Answer
可被公开 benchmark 的能力会被 AI 持续压缩。长期价值会迁移到更难被训练的地方：自发特质、判断力、真实世界权限、长期 good judgment，以及定义什么叫 Good 的权利。
Abilities that can be publicly benchmarked will keep being compressed by AI. Long-term value moves toward harder-to-train territory: self-generated traits, judgment, real-world permission, repeated good judgment, and the right to define what Good means.
关键方向
Key Directions
可测量就会被训练
Measurable Becomes Trainable
如果一件事可以被公开评分、反复比较、自动验证，它最终就会被 AI 学会或压缩。
If something can be publicly scored, repeatedly compared, and automatically verified, AI will eventually learn or compress it.
Untrainable 更像身份层
Untrainable Lives Closer To Identity
真正难训练的不是普通技能，而是没人监督时仍然学习、负责、发现问题并提高标准的自发特质。
The hardest things to train are not ordinary skills, but self-generated traits: learning without supervision, taking responsibility, noticing problems, and raising standards.
AI 负责 How，人负责 Why 和 What
AI Handles How; Humans Keep Why And What
AI 越来越擅长告诉你怎么做，但什么值得做、为什么要做、什么时候不做，仍然依赖 judgment。
AI gets better at explaining how to do things, but what is worth doing, why it matters, and when not to do it still depend on judgment.
长期价值来自定义 Good
Long-Term Value Comes From Defining Good
如果只追逐现有 benchmark，很容易被压缩。更深的位置，是通过长期判断和真实反馈，逐渐获得定义标准的权利。
If you only chase existing benchmarks, you are easy to compress. A deeper position is earning the right to define standards through repeated judgment and real-world feedback.
Mental Model Card
Mental Model Card
可训练性边界 Trainability Boundary
Trainability Boundary
能被公开测量的能力会被 AI 压缩；长期价值会迁移到判断、权限、真实世界反馈和定义标准的权利。
Abilities that can be publicly measured get compressed by AI; long-term value moves to judgment, permission, real-world feedback, and the right to define standards.
Core Idea
Core Idea
AI 时代不要只问“这个能力厉不厉害”，要问它能不能被公开评测、自动验证、规模化训练。如果能，它会越来越 trainable。更稀缺的部分在边界之外：发现什么重要、判断什么够好、承担责任，并持续形成 good judgment。
In the AI era, do not only ask whether a skill is impressive. Ask whether it can be publicly evaluated, automatically verified, and trained at scale. If yes, it becomes increasingly trainable. Scarcity moves beyond that boundary: noticing what matters, judging what is good enough, taking responsibility, and compounding good judgment.
Examples
Examples
- Coding benchmark 会被模型突破，但真实系统里的约束、责任边界和组织判断更难公开训练。Coding benchmarks can be beaten, but the constraints, accountability, and organizational judgment inside real systems are harder to train publicly.
- 专业领域里，长期价值往往来自对“够好”的判断，而不只是生成答案的速度。In professional domains, long-term value often comes from judging what is good enough, not merely producing answers quickly.
- 个人成长里，真正稀缺的是没人要求时仍然学习和提高标准。In personal growth, the scarce trait is continuing to learn and raise standards when nobody asks you to.
Applications
Applications
- 招聘时，不只问候选人会不会做，还要问没人告诉他时会不会主动发现并解决问题。When hiring, do not only ask whether someone can do the task; ask whether they notice and solve problems without being told.
- 做 AI 产品时，不只问模型能不能生成，还要问谁拥有真实反馈、permission 和 accountability。When building AI products, do not only ask whether the model can generate; ask who owns real feedback, permission, and accountability.
- 做个人知识系统时，每日问题要沉淀成可复用的判断模型，而不是停留在聊天记录。When building a personal knowledge system, daily questions should become reusable judgment models, not just chat logs.
一句值得记住的话
Quote
真正长期有价值的，不是赢下一次 benchmark，而是逐渐获得定义 benchmark 的权利。The durable advantage is not winning one benchmark; it is earning the right to define the benchmark.
讨论区
Discussion
欢迎用 GitHub 登录参与讨论。每篇文章会自动映射到独立的 GitHub Discussion。
Sign in with GitHub to join the conversation. Each post maps to its own GitHub Discussion.
Active Recall
Active Recall
- 为什么公开测量会让能力越来越 trainable？Why does public measurement make an ability increasingly trainable?
- Untrainable 为什么更像 self-generated traits？Why is the untrainable edge closer to self-generated traits?
- 一个产品如何从追逐 benchmark 走向定义 Good？How can a product move from chasing benchmarks to defining Good?