① 工程：你脑中那张"分布式系统应如何运作"的地图，和线上真实系统这片疆域，几乎所有事故都发生在两者分叉的缝隙处——"我以为这个接口是幂等的"。资深的标志不是地图画得更精确，而是时刻记得自己手里拿的只是地图，并把监控恰恰部署在这些缝隙上。② AI：用户画像、embedding 向量都是地图，真实的人是疆域；模型表现崩坏，常常是因为团队把地图当成了疆域，忘了它只在被训练的那个目的下才近似成立。
English Summary
Map Is Not the Territory — every concept, model, or intuition is a map; reality is the territory. Most error comes from mistaking the map for the territory itself. This isn't agnosticism: maps still differ in quality, but the test isn't "does it equal reality" (impossible) — it's "is it structurally similar to the territory, and useful for the current purpose?" Maps self-reinforce (you perceive through them, feedback is filtered by them), so the only escape is actively seeking where the map predicts wrongly, not where it holds. A good map is deliberately distorted for a purpose (the London Tube map sacrifices geography for connectivity). Mirrors the brain's predictive processing: perception is a reality-constrained hallucination shaped by your prior map.
AI Prompts
中文提示词
我对 [某领域/某系统/某人] 持有的核心判断是：[描述我脑中的"地图"]。请帮我做"地图 vs 疆域"审计：
① 这张地图最初是为了哪个目的画的？我现在是否在用它做超出该目的的判断？
② 指出 2-3 个最可能"地图与疆域分叉"的缝隙——我以为如此、实际可能不然的地方；
③ 给我 1 个可以主动验证"地图在哪里失效"的具体观察或实验。
English Prompt
My core belief about [domain/system/person] is: [describe my "map"]. Run a map-vs-territory audit:
1. What purpose was this map originally drawn for? Am I now using it to judge things beyond that purpose?
2. Name 2–3 likely gaps where map and territory diverge — places I assume X but reality may differ.
3. Give me one concrete observation or experiment that actively tests where the map breaks.
Hanlon's Razor — never attribute to malice what is adequately explained by stupidity (negligence, incompetence). It's a Bayesian prior-corrector: carelessness, miscommunication, and noise are statistically far more common than deliberate harm, so put your prior mass on "non-malice" first. The upgraded razor: often it's neither malice nor stupidity but a rational choice under constraints and incentives you can't see — invert the fundamental attribution error and explain behavior by situation, not character. Systems-thinking link: blaming a person often masks a systemic cause (bad process makes smart people fail). Boundary: the razor isn't an absolution — when "stupidity" forms a stable, self-serving pattern, update your prior; repeated "innocent mistakes" may be strategy.
AI Prompts
中文提示词
我遇到一件让我感觉被针对的事：[描述事件 + 我此刻的归因]。请用汉隆剃刀帮我重新审视：
① 列出 3 个"非恶意"的解释（疏忽 / 能力 / 沟通 / 随机）；
② 列出 1 个"约束与激励"解释——对方在什么处境下会理性地这么做？
③ 反向校验：有没有证据真的支持"恶意/策略"假设？若有，是哪些可复现的模式让我应当更新先验？
English Prompt
Something happened that feels aimed at me: [describe the event + my current attribution]. Apply Hanlon's Razor:
1. List 3 non-malice explanations (negligence / competence / miscommunication / randomness).
2. Give 1 "constraints & incentives" explanation — under what situation would the person rationally act this way?
3. Counter-check: is there real evidence for a malice/strategy hypothesis? If so, which reproducible pattern should make me update my prior?
① AI 超级个体：你可以用一群 agents 把执行带宽扩展到百倍，但能与你深度协作、彼此真实建模的"核心圈"仍受邓巴约束——工具放大的是产能，放大不了关系带宽。认清这条边界，才不会把精力浪费在维系一张根本撑不起来的网络上。② 精力管理：把人际投入按同心圆审计一遍——你是否在让外层的弱关系，悄悄吞掉了本应留给最里层 5 个人的时间？关系的复利只在最里两层才真正发生。
English Summary
Dunbar's Number — the human neocortex caps stable social relationships at roughly 150, structured as concentric circles (≈5 intimates, 15, 50, 150, 500, 1500), with bandwidth and depth decaying sharply outward. Relationship maintenance has a cognitive cost; past the cap, a relationship degrades from "an internal model of a real person" into a label. Social media creates pseudo-Dunbar inflation — you think you know 2000 people, but only ~150 are truly modeled; the rest are parasocial projections that drain attention without real reciprocity. Organizational corollary: past ~150, implicit "everyone-knows-everyone" trust can't scale coordination, forcing hierarchy and process — essentially the full-mesh complexity explosion of a social network. Practice: tier your social investment explicitly; attention is scarce, and maintaining the wrong layer carries a hidden opportunity cost.
AI Prompts
中文提示词
这是我当前的人际/协作网络现状：[描述我经常联系的人群、规模、精力分配]。请用邓巴数同心圆帮我审计：
① 把这些关系大致归入 5 / 15 / 50 / 150 层，指出哪一层被过度投入、哪一层被忽视；
② 找出 1-2 段"伪邓巴"关系——消耗注意力却缺乏真实互惠的拟社会关系；
③ 给出一个把精力重新分配回最里两层（复利层）的具体调整。
English Prompt
Here's my current relationship / collaboration network: [describe whom I contact regularly, the scale, my energy allocation]. Audit it with Dunbar's concentric circles:
1. Sort these relationships into the ≈5 / 15 / 50 / 150 layers; flag which layer is over-invested and which is neglected.
2. Identify 1–2 "pseudo-Dunbar" ties — parasocial relationships that drain attention without real reciprocity.
3. Propose one concrete reallocation that returns energy to the innermost two (compounding) layers.
哥德尔不完全性 · Gödel's Incompleteness
任何足够强的一致形式系统，都存在系统内为真却不可证的命题；且它无法在自身内部证明自身的一致性。 — Kurt Gödel, 1931
① 工程：一套测试套件永远无法证明自己"测全了"——谁来测试这些测试？系统内不存在该证明，所以你需要系统外的兜底：代码评审、形式化验证、混沌工程。把"测试通过"当成"代码正确"，就是把不完全的系统误当成完备。② 治理与自我认知：任何 KPI 体系最终都会被人钻空子（一种哥德尔式盲点），所以必须留一层人工判断来兜底；同理，你也无法仅用自己的思维框架去完整审计自己的思维框架——这正是他者反馈、冥想、反常数据不可替代的原因：它们是你思维系统之外的"元层"。
English Summary
Gödel's Incompleteness — any consistent formal system strong enough to express basic arithmetic is necessarily incomplete: there exist true-but-unprovable statements, and the system cannot prove its own consistency from within. Boundary first: this is a precise mathematical result, not a "nothing is knowable" mysticism. But its transferable core is rigorous: any self-consistent closed system has blind spots — it cannot verify all its own truths or self-certify its consistency using only its own rules. Transfers: (1) no rule system (law, KPIs, code standards) can pre-enumerate every case; boundary cases force a jump to a meta-level — human judgment, new axioms, external arbitration (Goodhart's Law is the practical version). (2) Self-referential verification has limits — you can't test a set of assumptions using those same assumptions. (3) AI link: a model can't judge the truth of its own output from within, which is why self-consistency is insufficient and external grounding plus human checking are required. Signal: when you're stuck unable to decide truth within a system, it's time to ascend a meta-level rather than grind harder inside it.
AI Prompts
中文提示词
我正在依赖这个系统来做判断/治理/验证：[描述系统——KPI 体系 / 测试套件 / 规则集 / 自我反思流程]。请用哥德尔式的"系统盲点"视角分析：
① 这个系统最可能在哪类"边界案例"上无法在内部判定？
② 它有没有试图"自证一致"的地方——用自身规则验证自身？指出风险；
③ 给出 1 个具体的"元层"兜底：引入什么外部视角 / 新前提 / 他者裁决，能补上这个盲点。
English Prompt
I rely on this system to judge / govern / verify: [describe the system — a KPI scheme / test suite / rule set / self-reflection process]. Analyze it through Gödel's "blind spot" lens:
1. On what kind of boundary cases is this system most likely unable to decide from within?
2. Where does it try to "self-certify consistency" — verify itself using its own rules? Name the risk.
3. Propose one concrete meta-level backstop: which external perspective / new premise / outside arbitration would patch the blind spot.