把全部积蓄和所有业余时间押在单一 AI 创业点子上,即使它期望为正,也违反凯利——归零风险会让你的人生几何增长崩盘。正确做法是按「优势大小」分配:在你真正有信息差的领域(AI、分布式)多配,在你只是跟风的领域少配或不配,且永远留出不被单次失败清零的缓冲。职业、健康、育儿都是重复博弈,任何一项 all-in 都是反凯利的。
English Summary
Kelly Criterion — in repeated bets you should maximize long-run geometric growth (log-wealth), not single-shot arithmetic expectation. Optimal bet fraction = edge ÷ odds; the more your edge, the more you bet; the more variance, the less. Full-betting any positive-EV wager with nonzero ruin probability leads almost surely to bankruptcy, because ruin is an absorbing state. Use half-Kelly: people systematically overestimate their edge, and overbetting is punished asymmetrically — slight underbetting costs little, slight overbetting wrecks growth. Kelly equals Shannon's information rate: your edge is your information advantage; with no real edge, the optimal stake is zero.
AI Prompts
中文提示词
我面对一个重复性的决策/下注:[描述机会、估计胜率、赔率/上下行]。请用凯利框架帮我:
① 估算这是正期望还是负期望,最优下注比例大概多少;
② 评估我对「优势」的估计可能高估在哪,建议一个保守的「半凯利」仓位;
③ 指出这个赌注里有没有「归零/不可逆」的尾部,若有,仓位该如何再下调。
English Prompt
I face a repeated decision/bet: [describe the opportunity, your win probability, odds/upside-downside]. Use the Kelly framework to:
1. Estimate whether it's positive- or negative-EV and the rough optimal bet fraction.
2. Stress-test where I'm likely overestimating my edge, and suggest a conservative half-Kelly size.
3. Flag any ruin / irreversible tail in this bet; if present, advise how much further to cut the size.
Convex Betting (Optionality) — the point isn't predicting correctly but engineering a payoff shape: capped downside, open-ended upside. Such positions are convex, so by Jensen's inequality volatility itself creates value — the wilder the swings, the higher the expected payoff. You needn't know which bet pays off, only that each is "affordable to lose, unbounded to win." The opposite is concave: many "safe" strategies are actually picking up pennies in front of a steamroller. Use a barbell: mostly ultra-safe, a small slice on extreme convex bets, avoiding the false-safe middle. Same structure as biological mutation — cheap convex trials, returns carried by rare fat-tailed wins.
AI Prompts
中文提示词
我正在考虑投入 [项目/赌注/资源配置]。请用凸性视角帮我审视:
① 它的收益形状是凸(下行封顶、上行开放)还是凹(收益封顶、灾难无底)?
② 如果是凹的,怎么改造成凸的——把损失截断在哪、让上行如何打开?
③ 帮我设计一个「杠铃」配置:哪部分该极度保守,哪一小部分该押凸性厚尾。
English Prompt
I'm considering committing to [project/bet/resource allocation]. Examine it through a convexity lens:
1. Is its payoff shape convex (capped downside, open upside) or concave (capped gains, bottomless disaster)?
2. If concave, how could I reshape it into convex — where to cap the loss, how to open the upside?
3. Help me design a barbell: which part should be ultra-safe, and which small slice should bet on convex fat tails.
Tail-Risk Hedging — the mirror image of convex betting: pay a small, long-run negative-EV premium to buy convex protection against rare, devastating events. The goal isn't profit but never being zeroed out — staying at the table to enjoy compounding. The key is ergodicity: a positive group-average return doesn't mean an individual survives over time, because ruin is an irreversible absorbing state. Distinguish recoverable losses from ruin — hedge only the latter; bear volatility on the former. It's psychologically brutal: people hate steady small premiums more than they fear rare catastrophe, so they cancel coverage right before the crash. Ask first "what could take me out of the game," then insure that.
AI Prompts
中文提示词
我的处境/系统是 [描述]。请帮我做一次「灾难尾」体检:
① 列出 3 个最可能让我「彻底出局/不可逆破产」的尾部事件;
② 区分哪些损失是可恢复的(该承受波动)、哪些是不可恢复的(该对冲);
③ 为不可恢复的那几个,给出成本最低的对冲/冗余方案,并说明保费值不值。
English Prompt
My situation/system: [describe]. Run a tail-risk check-up:
1. List the 3 tail events most likely to take me out of the game / cause irreversible ruin.
2. Separate recoverable losses (bear the volatility) from unrecoverable ones (hedge them).
3. For the unrecoverable ones, propose the lowest-cost hedge/redundancy, and justify whether the premium is worth it.
Insurance Thinking — the right use of insurance is to transfer risks that could take you out of the game, not affordable risks you can self-fund. Most people buy it backwards: paying premiums on phone-screen or appliance warranties (negative-EV, affordable — should self-insure) while going bare on illness, liability, or income loss (low-probability, ruinous — should transfer). Insurance is negative-EV but positive-utility because utility is concave in wealth: losing a fortune hurts more than gaining one delights, so paying a little to shave the left tail is rational. Self-insuring requires a buffer — your reserves are your own insurance company. Watch for moral hazard: hedging changes behavior, so align incentives via skin in the game.
AI Prompts
中文提示词
这是我目前的风险与保险配置:[列出现有的保险/缓冲、主要担忧]。请用保险思维帮我重排:
① 把我的风险摆进「频率 × 损失大小」四象限,标出哪些该自保、哪些该投保;
② 找出我「为小确定性过度付费」和「对大尾部裸奔」的地方;
③ 给出一个调整方案:先建多少现金缓冲,再把哪几个不可承受的尾部转移出去。
English Prompt
Here's my current risk/insurance setup: [list existing coverage/buffer, main worries]. Rebalance it with insurance thinking:
1. Place my risks on a frequency × severity quadrant; mark what to self-insure vs transfer.
2. Find where I'm overpaying for small certainties and where I'm bare against large tails.
3. Propose an adjustment: how much cash buffer to build first, then which unbearable tails to transfer out.