机器学习里的生成对抗网络（GAN）是同一套哲学的数学化：生成器拼命造假、判别器拼命识假，两者在对抗中共同逼出以假乱真的结果——没有哪一方"知道"真相，真相是对抗的产物。同理还有代码评审（作者 vs 评审者）、AI 安全里的红队、乃至用"辩论"做对齐。对 BigCat：当你不确定一个方案是否可靠，与其找一个"权威"拍板，不如制造一场有裁判的对抗——让最锋利的反方来攻击它，扛得住攻击的才可信。
English Summary
Adversarial vs Cooperative — two machines for reaching truth. The adversarial system (common law) sets two opposed parties each pushing their side to the limit while a neutral arbiter decides; truth emerges from rule-bound conflict. The inquisitorial/cooperative system (civil law) has the judge actively investigate. The deep difference lies in their assumption about people: the adversarial design assumes everyone is biased and self-serving, so instead of trusting any single party it institutionalizes the biases and lets them cancel out. Non-trivial: (1) truth can be engineered out of structured opposition rather than relying on a trustworthy saint — you need not a wise judge but two oppositely-motivated parties plus rules. (2) Its hidden pillar is meta-level neutrality and rules; capture the referee or drop the rules and it degrades into a brawl where the loudest or best-resourced wins, burying truth deeper. (3) Use adversarial when parties may act in bad faith (mutual policing); use cooperative when goals are genuinely shared (far less overhead). Key terms: adversarial system, inquisitorial system, neutral arbiter, structured conflict.
AI Prompts
中文提示词
我要判断一个方案／结论是否可靠：[描述这个方案或结论]。请帮我用「对抗式」逼出它的真实成色：
① 扮演最锋利的反方，用尽全力攻击它，找出最致命的漏洞；
② 再扮演正方为它辩护，看哪些攻击能被挡住、哪些挡不住；
③ 作为中立裁判，基于这场对抗给出结论：它在哪些条件下可信、哪些条件下会崩。
English Prompt
I need to judge whether a plan or conclusion is sound: [describe the plan or conclusion]. Use an adversarial process to reveal its true quality:
1. Play the sharpest opposing counsel and attack it with everything, surfacing its most fatal flaws.
2. Then play its defender and see which attacks can be parried and which cannot.
3. As a neutral arbiter, rule on the basis of this clash: under what conditions it holds, and under what conditions it breaks.
Default Rules — contract law distinguishes default rules (apply unless parties contract around them) from mandatory rules (cannot be waived). Most law is default rules, filling the gaps. The technical-looking distinction hides a force that governs reality: because almost no one ever changes the default, the default is effectively the rule. Opt-in vs opt-out organ donation yields wildly different rates among near-identical populations; auto-enrollment doubles retirement saving. Non-trivial: (1) defaults rule through inertia, status-quo bias, and the endorsement effect ("the system chose it, so it's probably right") — they are the path of least resistance, and people follow it. (2) Two opposite design philosophies: majoritarian defaults (set to what most want, minimizing transaction costs) vs penalty defaults (deliberately set to what nobody wants, forcing the informed party to speak up and reveal private information). (3) Therefore your default is your policy — 95% never move it; "I offered a choice" is an illusion, what you offered was the default. Key terms: default vs mandatory rules, majoritarian default, penalty default, status-quo bias.
AI Prompts
中文提示词
我想改变某个群体／用户的行为，或正在设计一个系统的设定：[描述场景与目标行为]。请用「默认规则」帮我：
① 找出当前的默认值是什么、它实际把人导向了哪里（而非我以为的"选项自由"）；
② 给出"多数派默认"方案：把默认设成多数人本就想要的，降低摩擦；
③ 再给出"惩罚性默认"方案：哪个环节该故意设成必须显式选择，以逼出关键信息或防止无意识犯错。
English Prompt
I want to change a group's or users' behavior, or I'm designing a system's settings: [describe the situation and the target behavior]. Use default rules to help me:
1. Identify the current default and where it actually steers people (not the illusory freedom of "options").
2. Propose a majoritarian default: set it to what most people already want, reducing friction.
3. Propose a penalty default: which step should deliberately require an explicit choice, to force out key information or prevent unconscious mistakes.
AI 对齐问题，本质就是一份无法写完的契约：你无法把"我到底想要什么"完整地写进奖励函数，模型便钻进你没写到的缝隙——奖励黑客、规格博弈（specification gaming）。人类意图与模型目标之间的缺口，是不可消除的"契约不完全"。软件世界同理：API 永远无法规定全部行为，于是海勒姆定律说"所有可观察行为终将被人依赖"，未定义行为就是合同的留白。结论是一致的：与其幻想写出完备规格，不如设计好"缺口出现时由谁、按何机制裁断"。
English Summary
Incomplete Contracts — a Nobel-winning insight: every contract is necessarily incomplete. The world is too complex, the future not fully foreseeable, and specifying every contingency is prohibitively costly — so all real contracts leave gaps. Once you accept this, the key question shifts from "how to write the perfect contract" to "who decides in the situations it didn't cover." Non-trivial: (1) this is the answer to why firms exist — ownership is residual control rights over contractually-unspecified situations; you acquire a company not because you can write everything down but to hold the deciding power where the contract is silent. (2) Chasing a "complete contract" is both impossible and harmful — brittle, endless, and a signal of distrust; experts don't pile up clauses but design gap-filling mechanisms: governance, long-term relationships, room to renegotiate. Trust and repeated games fill the holes. (3) The more long-term and unforeseeable the relationship, the thinner the contract and the thicker the mechanism should be. Key terms: incompleteness, residual control rights, theory of the firm, relational contracts.
AI Prompts
中文提示词
我正在拟定一份协议、规范或制度，想把它写得尽量周全：[描述这份契约和它要管的关系]。请用「不完全契约」帮我换个重心：
① 指出哪些情形我根本无法事先穷尽预见（缺口在哪）；
② 与其堆条款，不如设计"缺口出现时由谁来决定、按什么程序"——给出剩余控制权与治理机制的安排；
③ 评估这段关系该用"薄合同 + 厚机制"还是"厚合同"，并说明理由。
English Prompt
I'm drafting an agreement, spec, or policy and want it as watertight as possible: [describe the contract and the relationship it governs]. Use incomplete contracts to shift my focus:
1. Point out which contingencies I simply cannot foresee in advance (where the gaps are).
2. Instead of piling up clauses, design who decides when a gap appears, and by what procedure — an arrangement of residual control rights and governance.
3. Judge whether this relationship wants a "thin contract + thick mechanism" or a "thick contract," and explain why.
Chesterton's Fence — Chesterton's parable: a fence stands inexplicably across a road; a reformer says "I see no use in it, tear it down." The wise reply: "Precisely because you don't see its use, I won't let you remove it — go find out why it's there, and then come back to remove it." It is an iron rule of epistemic humility: a thing's lack of an obvious reason is first evidence of your ignorance, not of its uselessness. Non-trivial: (1) it counters the reformer's arrogance — rules, structures, and traditions that survived were often selected for reasons invisible to you now (kin to the Lindy effect and evolutionary selection). (2) It is not conservatism for its own sake — once you truly understand the fence's origin, you are entitled, even obliged, to remove it if that reason no longer holds; it constrains not whether you may change, but the order of operations: understand first, then act. (3) It guards against second-order effects — the fence's function often reveals itself only after you remove it and the cattle escape. Key terms: epistemic humility, second-order effects, Lindy effect, order of operations.
AI Prompts
中文提示词
我想删除／改掉一个看起来多余或过时的东西（一段代码、一条规则、一个流程、一项传统）：[描述它]。请用「切斯特顿栅栏」拦住我的冲动：
① 先别评价该不该拆——帮我重建它当初被立起来的可能理由（哪怕现在看不出）；
② 列出如果贸然拆掉，可能爆发的二阶效应与隐藏依赖；
③ 给出判断：那个最初的理由今天还成立吗？若已消失，再告诉我可以放心拆。
English Prompt
I want to delete or change something that looks redundant or outdated (a piece of code, a rule, a process, a tradition): [describe it]. Use Chesterton's Fence to check my impulse:
1. Before judging whether to remove it, help me reconstruct the reasons it may have been put there in the first place (even if invisible now).
2. List the second-order effects and hidden dependencies that could erupt if I remove it rashly.
3. Give a verdict: does that original reason still hold today? If it has vanished, then tell me I can safely remove it.