我的征尘是星辰大海。。。
The dirt and dust from my pilgrimage forms oceans of stars...
-------当记忆的篇章变得零碎，当追忆的图片变得模糊，我们只能求助于数字存储的永恒的回忆
作者:黄教授
手机视频列表
靠谱不靠谱忽悠更靠谱法诺不等式给我们的启示
视频
音频
原始脚本
为什么马拉多纳平球比贝利更有用？从信息论的法诺不等式看懂靠谱的真正含义。 看球时总有人纠结，该信贝利的预测，还是马拉多纳的？贝利好歹有70%的正确率，马拉多纳10次能错9次。 可真要下注，反着马拉多纳来，赢的概率反而更高。 这不是随口调侃的反向指标，背后藏着信息论里一个经典工具，法诺不等式，Fano's inequality。 它能帮我们精准判断谁的建议更有用，甚至能解释为什么战略忽悠也有实实在在的价值。 先搞懂法诺不等式是干嘛的，不用怕公式，看核心就行。 先给这个理论证个名，法诺不等式是信息论领域的基础定理之一，由美国科学家罗伯特法诺在1950年代提出。 它的核心作用特别朴素，就解决一个问题，当我们获得某个信息，比如专家的预测，一个信号后，对事情真相的不确定感最多能减少多少？简单说，它就像一把信息价值尺子。 不管是专家预测、天气预报，还是朋友给的建议，只要能算出这个信息的规律有多稳定，用这把尺子一量，就知道它能不能帮我们做决定。 而且结论往往和正确率高低没那么直接。 回到例子，贝利 vs 马拉多纳，尺子量出谁更有用。 我们不用算复杂的对数公式，就用这把尺子量两个球迷熟悉的专家。 专家 A，贝利，70%正确率，模糊正确派。 贝利预测世界杯10次里有7次对，听着靠谱。 但法诺不等式会告诉我们，他的对不够稳定。 你每次听他说巴西队赢，虽然有70%的概率压对，但心里总悬着，这次会不会是那30%错的？用尺子量的结果是，他的预测只能帮你把不确定感从1比特完全瞎猜，降到0.88比特。 减少的幅度很小。 简单说，你还是会犹豫，因为规律不够清晰。 专家 B，马拉多纳，10%正确率，稳定错误派。 马拉多纳刚好相反，10次预测9次错，但他的错特别规律，只要他说 A 队赢，你压 B 队赢，10次里能中9次。 法诺不等式亮出来的结果更惊人，他的预测能把不确定感直接降到0.469比特，几乎减少了一半还多，为什么？吗？因为90%错的规律太明确了，你不用纠结要不要反着来，决策时心里特别有底。 关键结论，法诺不等式告诉我们靠谱的真正定义。 很多人以为有用就是正确率高，但法诺不等式戳破了这个误区。 信息的价值核心是规律的稳定性，不是正确率的高低。 就像我们说张召忠的战略忽悠有用，不是因为他说对了多少，而是因为他的判断有稳定的偏向。 如果他总能精准说错，那这种稳定的错反而比时对时错的建议更有用。 法诺不等式的本质就是帮我们抓住规律这个核心，哪怕是错的规律，只要稳定就能转化成我们的决策依据。 再举个生活里的例子，你问同事今天会不会堵车。 同事 A 说可能堵也可能不堵，没规律。 同事 B 说每次说堵都不堵，稳定错。 按法诺不等式的逻辑，同事 B 的建议反而更有用。 你直接反着他的话安排出门，比听同事 A 的模棱两可靠谱多了。 最后，下次听建议先问有规律吗？再问对不对？我们平时找靠谱的人，其实找的不是总对的人，而是有稳定规律的人。 法诺不等式给我们的启示特别简单，不管是专家预测、职场建议，还是生活里的小判断。 先别问他以前对过多少次，先观察他的判断有规律吗？哪怕是总错的规律，只要稳定，就比时对时错的模糊正确更有价值。 因为前者能帮你减少不确定感，后者只会让你更纠结。 下次再看球下注，听专家分析，不妨在心里用法诺不等式这把尺子量一量。 他的话有规律吗？能让我少一点犹豫吗？想清楚这两个问题，比纠结正确率多少管用多了。
修正脚本
为什么马拉多纳评球比贝利更有用？从信息论的法诺不等式看懂靠谱的真正含义。 看球时总有人纠结，该信贝利的预测，还是马拉多纳的？贝利好歹有70%的正确率，马拉多纳10次能错9次。 可真要下注，反着马拉多纳来，赢的概率反而更高。 这不是随口调侃的反向指标，背后藏着信息论里一个经典工具，法诺不等式，Fano's inequality。 它能帮我们精准判断谁的建议更有用，甚至能解释为什么战略忽悠也有实实在在的价值。 先搞懂法诺不等式是干嘛的，不用怕公式，看核心就行。 先给这个理论证个名，法诺不等式是信息论领域的基础定理之一，由美国科学家罗伯特法诺在1950年代提出。 它的核心作用特别朴素，就解决一个问题，当我们获得某个信息，比如专家的预测，一个信号后，对事情真相的不确定感最多能减少多少？简单说，它就像一把信息价值尺子。 不管是专家预测、天气预报，还是朋友给的建议，只要能算出这个信息的规律有多稳定，用这把尺子一量，就知道它能不能帮我们做决定。 而且结论往往和正确率高低没那么直接。 回到例子，贝利 vs 马拉多纳，尺子量出谁更有用。 我们不用算复杂的对数公式，就用这把尺子量两个球迷熟悉的专家。 专家 A，贝利，70%正确率，模糊正确派。 贝利预测世界杯10次里有7次对，听着靠谱。 但法诺不等式会告诉我们，他的对不够稳定。 你每次听他说巴西队赢，虽然有70%的概率压对，但心里总悬着，这次会不会是那30%错的？用尺子量的结果是，他的预测只能帮你把不确定感从1比特完全瞎猜，降到0.88比特。 减少的幅度很小。 简单说，你还是会犹豫，因为规律不够清晰。 专家 B，马拉多纳，10%正确率，稳定错误派。 马拉多纳刚好相反，10次预测9次错，但他的错特别规律，只要他说 A 队赢，你压 B 队赢，10次里能中9次。 法诺不等式亮出来的结果更惊人，他的预测能把不确定感直接降到0.469比特，几乎减少了一半还多，为什么？因为90%错的规律太明确了，你不用纠结要不要反着来，决策时心里特别有底。 关键结论，法诺不等式告诉我们靠谱的真正定义。 很多人以为有用就是正确率高，但法诺不等式戳破了这个误区。 信息的价值核心是规律的稳定性，不是正确率的高低。 就像我们说张召忠的战略忽悠有用，不是因为他说对了多少，而是因为他的判断有稳定的偏向。 如果他总能精准说错，那这种稳定的错反而比时对时错的建议更有用。 法诺不等式的本质就是帮我们抓住规律这个核心，哪怕是错的规律，只要稳定就能转化成我们的决策依据。 再举个生活里的例子，你问同事今天会不会堵车。 同事 A 说可能堵也可能不堵，没规律。 同事 B 说每次说堵都不堵，稳定错。 按法诺不等式的逻辑，同事 B 的建议反而更有用。 你直接反着他的话安排出门，比听同事 A 的模棱两可靠谱多了。 最后，下次听建议先问有规律吗？再问对不对？我们平时找靠谱的人，其实找的不是总对的人，而是有稳定规律的人。 法诺不等式给我们的启示特别简单，不管是专家预测、职场建议，还是生活里的小判断。 先别问他以前对过多少次，先观察他的判断有规律吗？哪怕是总错的规律，只要稳定，就比时对时错的模糊正确更有价值。 因为前者能帮你减少不确定感，后者只会让你更纠结。 下次再看球下注，听专家分析，不妨在心里用法诺不等式这把尺子量一量。 他的话有规律吗？能让我少一点犹豫吗？想清楚这两个问题，比纠结正确率多少管用多了。
英文翻译
Why is Maradona's commentary more useful than Pelé's? Understanding the true meaning of reliability through Fano's inequality from information theory. When watching games, people often struggle over whether to trust Pelé's predictions or Maradona's. Pelé at least has a 70% accuracy rate, while Maradona gets 9 out of 10 wrong. But when it comes to actually placing bets, going against Maradona gives you a higher probability of winning. This isn't just a casual joke about a reverse indicator; behind it lies a classic tool from information theory: Fano's inequality. It helps us precisely determine whose advice is more useful, and can even explain why strategic deception (like "strategic忽悠") has tangible value. First, let's understand what Fano's inequality does. Don't be afraid of the formula; just focus on the core idea. To give this theory its due, Fano's inequality is a fundamental theorem in information theory, proposed by American scientist Robert Fano in the 1950s. Its core function is very simple: it addresses one question—when we obtain a piece of information, such as an expert's prediction or a signal, what is the maximum amount we can reduce our uncertainty about the truth? In simple terms, it's like a ruler for measuring information value. Whether it's expert predictions, weather forecasts, or advice from a friend, as long as we can calculate how stable the pattern of this information is, using this ruler tells us whether it can help us make decisions. And the conclusion often isn't directly related to how high the accuracy rate is. Back to the example: Pelé vs. Maradona. The ruler measures who is more useful. We don't need to calculate complex logarithmic formulas; we just use this ruler on two experts familiar to fans. Expert A: Pelé, 70% accuracy, the "vague correctness" school. Pelé gets 7 out of 10 World Cup predictions right, which sounds reliable. But Fano's inequality tells us his correctness isn't stable enough. Every time you hear him say Brazil will win, although there is a 70% chance of being correct, you still worry: what if this is the 30% that's wrong? The ruler's result is that his prediction only reduces your uncertainty from 1 bit (completely guessing) to 0.88 bits. The reduction is very small. Simply put, you still hesitate because the pattern isn't clear enough. Expert B: Maradona, 10% accuracy, the "stable error" school. Maradona is the opposite: 9 out of 10 predictions are wrong, but his errors are very consistent. Whenever he says Team A will win, you bet on Team B, and you win 9 out of 10 times. Fano's inequality reveals an even more surprising result: his prediction can directly reduce uncertainty to 0.469 bits—a reduction of more than half. Why? Because the pattern of 90% errors is so clear that you don't have to wrestle with whether to go against him; your decision becomes much more confident. Key conclusion: Fano's inequality tells us the true definition of reliability. Many people think usefulness means high accuracy, but Fano's inequality shatters this misconception. The core of information value is the stability of the pattern, not the level of accuracy. Just as we say Zhang Zhaozhong's "strategic deception" is useful, it's not because he gets many things right, but because his judgments have a stable bias. If he can consistently be wrong in a precise way, then that stable wrongness is more useful than advice that is sometimes right and sometimes wrong. The essence of Fano's inequality is to help us grasp this core of pattern. Even if the pattern is wrong, as long as it's stable, it can be transformed into a basis for our decisions. Let's take another example from daily life: you ask a colleague whether today's commute will be congested. Colleague A says it might be congested or not—no pattern. Colleague B says every time he says it's congested, it's not—stable error. According to Fano's inequality, Colleague B's advice is actually more useful. You simply do the opposite of what he says when planning your departure, which is far more reliable than listening to Colleague A's vague and ambiguous statements. Finally, next time you hear advice, first ask: "Is there a pattern?" then ask: "Is it right or wrong?" When we look for reliable people, we aren't looking for someone who is always right, but someone with a stable pattern. The insight from Fano's inequality is very simple: whether it's expert predictions, workplace advice, or small judgments in life. Don't first ask how many times they were right before; first observe whether their judgments have a pattern. Even if it's a pattern of always being wrong, as long as it's stable, it's more valuable than vague correctness that is sometimes right and sometimes wrong. Because the former helps reduce your uncertainty, while the latter only makes you more indecisive. Next time you watch a game and place a bet, or listen to expert analysis, mentally use Fano's inequality as a ruler to measure. Does what they say have a pattern? Can it make me hesitate less? Thinking through these two questions is far more useful than obsessing over accuracy rates.
back to top