① RLHF 的本质就是一道阿罗难题——把成千上万标注者两两比较的偏好，聚合成「一个」奖励模型。有用、无害、诚实这几个维度上的排序天然打架，不存在无损的聚合，所谓「对齐」永远是在公理之间取舍。② 团队用投票给需求排优先级，常陷入孔多塞循环：换个投票顺序就换个结论。识破之后，与其反复投票，不如明确决策维度、由负责人定锚——这正是「非独裁」做不到时的务实退路。
English Summary
Arrow's Impossibility Theorem — with 3+ options and 3+ voters, no preference-aggregation rule can simultaneously satisfy a handful of minimal fairness axioms (unanimity/Pareto, independence of irrelevant alternatives, non-dictatorship). A perfect way to turn individual preferences into a collective ranking simply doesn't exist; every voting or scoring system silently violates one axiom. The intuitive core is the Condorcet paradox: rational individuals can produce an intransitive collective preference that cycles (A>B>C>A) — rationality evaporates in aggregation. Because independence of irrelevant alternatives fails in practice (the spoiler effect), real power lies in agenda-setting, not voting. Modern instance: RLHF is an Arrow problem — aggregating thousands of pairwise human preferences into one reward model across conflicting dimensions has no lossless solution, so "alignment" is always a trade-off among axioms.
AI Prompts
中文提示词
我正在用 [投票 / 打分 / 排序 / 多目标聚合] 的方式做一个集体决策：[描述场景与候选项]。请用「阿罗不可能定理」帮我看清：
① 我现在的聚合规则，悄悄牺牲了哪条公平原则（全体一致 / 无关选项独立 / 非独裁）？
② 这里有没有孔多塞循环或搅局者效应的风险？换个投票顺序，结论会不会变？
③ 既然完美聚合不存在，给我 1 个务实方案：要么明确锚定决策维度，要么把取舍摊开让所有人知情。
English Prompt
I'm making a collective decision via [voting / scoring / ranking / multi-objective aggregation]: [describe the setup and options]. Use Arrow's Impossibility Theorem to clarify:
1. Which fairness axiom is my current aggregation rule silently violating (unanimity / independence of irrelevant alternatives / non-dictatorship)?
2. Is there a risk of a Condorcet cycle or spoiler effect — would changing the voting order change the outcome?
3. Since no perfect aggregation exists, give me one pragmatic move: either anchor the decision on explicit dimensions, or surface the trade-off so everyone decides informed.
① 一份共享的工程代码库就是典型公地：人人图快往里塞补丁（私人收益：早点交付），没人愿意重构（公共成本：整体腐化）。技术债的本质就是一场缓慢的公地悲剧。解法不是骂人不自觉，而是建结构：明确 owner、把重构计入排期、用 CI 门槛设「逐级惩罚」。② 全社会的注意力也是一块公地，正被推荐算法争相开采——每个 app 多抢你三秒，合起来把你的专注力耗成碎片。③ 家里的共享空间、共担家务同理：与其指望自觉，不如和孩子一起立「清晰边界 + 轮值规则」，这正是公共治理的家庭微缩版。
English Summary
Tragedy of the Commons — a finite, shared, non-excludable resource gets exhausted because each user rationally takes a bit more while the depletion cost is spread across everyone: private gain, socialized cost. Every herder knows the pasture will degrade, but "if I don't graze more, someone else will," so all overgraze until ruin. The commons is destroyed not by bad people but by structure — swap in saints and the outcome is the same, so the fix must be structural, not moral exhortation. "Privatize or let the state regulate" is a false binary: communities can self-govern a commons sustainably through clear boundaries, low-cost monitoring, and graduated sanctions (Ostrom). Its mirror image is the anticommons — too many veto-holders leave a resource unused. The core crack is always the same: those who benefit and those who bear the cost are different people (an externality).
AI Prompts
中文提示词
我面对一个共享资源被过度消耗的局面：[描述这块「公地」——代码库 / 注意力 / 预算 / 公共空间等]。请用「公地悲剧」帮我分析：
① 这里的「私人收益、全体成本」裂缝具体长什么样？谁在多占，成本被摊给了谁？
② 别从道德入手——给我 2 个结构性解法（清晰边界 / 低成本监督 / 逐级惩罚 / 明确 owner），让可持续成为理性选择。
③ 有没有矫枉过正、滑向「反公地悲剧」（太多人握有否决权、资源闲置）的风险？
English Prompt
I'm facing a shared resource being overused: [describe the "commons" — codebase / attention / budget / shared space]. Use the Tragedy of the Commons to analyze:
1. What exactly is the "private gain, collective cost" crack here? Who is over-taking, and onto whom is the cost spread?
2. Skip the moralizing — give me 2 structural fixes (clear boundaries / low-cost monitoring / graduated sanctions / a named owner) that make sustainability the rational choice.
3. Is there a risk of over-correcting into an "anticommons" (too many veto-holders, resource left idle)?
① 开源项目是活标本：极少数维护者扛起百万使用者的依赖，绝大多数人只索取、不回馈。它能续命，全靠「选择性激励」——声誉、履历、招聘机会这些专属于贡献者的私人回报。② 团队复盘里人人都说「构建太慢、该治理了」，散会却没人动手——典型的集体行动困境。别再开会呼吁，要给私人钩子：指定 owner、把它写进某人的 OKR、公开表彰修好的人。③ 想在学校组织家长协作，也别只喊「为了孩子」，得设计专属激励，否则人人观望。
English Summary
The Logic of Collective Action (Olson) — shared interest does NOT automatically produce shared action. Even when a whole group clearly benefits from a goal, rational individuals often won't pitch in, because the payoff is shared (public) while the effort is private; everyone waits for "someone else" and no one acts. The larger the group, the weaker each member's incentive — my contribution is negligible and I get the benefit anyway. This yields an iron rule: small, concentrated interests beat large, diffuse ones (policy capture by lobbies). Olson's remedy is selective incentives — you can't mobilize people on the collective good alone; you must attach a private reward or punishment exclusive to participants (member perks, status, public recognition). A counterintuitive corollary, "exploitation of the great by the small": the big player who values the good most ends up footing the bill while small members free-ride.
AI Prompts
中文提示词
有一件大家都认同该做、却迟迟没人动手的集体好事：[描述这件事和涉及的群体]。请用「集体行动的逻辑」帮我破局：
① 拆解这里「收益公共、成本私人」的结构，以及群体规模如何稀释了每个人的动机。
② 别再靠呼吁共同利益——给我 2 个「选择性激励」设计：用专属于参与者的声望 / 福利 / 点名（正向）或代价（负向）把人撬动起来。
③ 是不是存在「小人物剥削大人物」——某个最在乎的人正被迫独自买单？该如何重新分摊？
English Prompt
There's a collective good everyone agrees should happen but no one acts on: [describe it and the group involved]. Use the Logic of Collective Action to break the deadlock:
1. Unpack the "public payoff, private cost" structure, and how group size dilutes each person's incentive.
2. Stop appealing to shared interest — give me 2 selective-incentive designs: use reputation / perks / public recognition (positive) or costs (negative) exclusive to participants to mobilize people.
3. Is there "exploitation of the great by the small" — one person who cares most being forced to foot the bill alone? How should the load be redistributed?
① 团队里的共享文档与知识就是块公共品：写文档是私人成本，好处大家白拿，于是文档长期欠奉、知识困在个人脑里。解法是制造「排他感」——把写文档计入考核、给作者署名与声望。② 给开源模型、公共数据集做贡献也是同一困境：你的算力是私人成本，能力红利却归全人类。③ 连家里也有：谁来维护那本共享日历、谁记录孩子的成长档案？看清它是公共品，就别再等「自然有人做」，要明确指派、给出专属认可。
English Summary
The Public Goods Paradox — public goods have two defining traits: non-excludability (you can't stop non-payers from using them) and non-rivalry (my use doesn't diminish yours) — lighthouses, clean air, defense, knowledge. These two traits make markets systematically under-provide them: you can't charge beneficiaries, and since one more user costs ~zero the efficient price is zero anyway — so who pays? Everyone waits for someone else, and the good everyone wants goes unbuilt. The paradox: the more universally desired a good is, the scarcer it becomes precisely because everyone wants it for free. Two failure engines: non-excludability kills the incentive to charge; non-rivalry means the optimal price is zero (hence no revenue). That's why we need governments, patronage, or clever mechanisms (assurance contracts, matching funds). Information and knowledge are the ultimate public goods — digital non-rivalry is near-total (infinite copies, ~zero marginal cost), so the real modern battle is manufacturing artificial excludability (DRM, paywalls, API keys), the deep tension under open source, public science, and AI.
AI Prompts
中文提示词
我面对一个「大家都想要、却没人愿意创造 / 维护」的东西：[描述这个公共品——团队文档 / 共享工具 / 开源贡献 / 家庭事务]。请用「公共品悖论」帮我分析：
① 它的「非排他、非竞争」具体体现在哪？为什么会陷入人人白嫖、无人供给？
② 给我 2 个供给方案：如何人为制造「排他感」或协调付费（计入考核 / 署名声望 / 达标才启动的众筹式承诺 / 配捐）？
③ 谁是那个最该被激励去供给它的关键人？专属回报该怎么给？
English Prompt
I'm facing something everyone wants but no one will create/maintain: [describe the public good — team docs / shared tooling / open-source contribution / household task]. Use the Public Goods Paradox to analyze:
1. How do non-excludability and non-rivalry show up here, and why does it collapse into everyone free-riding and no one providing?
2. Give me 2 provisioning fixes: how to manufacture "excludability" or coordinate payment (tie it to evaluation / authorship & reputation / an assurance-contract-style "only if enough commit" / matching funds)?
3. Who is the key person most worth incentivizing to provide it, and what exclusive reward should they get?