优化胜率而非赔率，把一件事做到理论上该有的样子

优化胜率而非赔率，把一件事做到理论上该有的样子Optimize for win rate, not odds — do things the way they theoretically should be done

42章经42 Chapters关注Follow

本文来自微信公众号：42章经，作者：曲凯，原文标题：《优化胜率而非赔率，把一件事做到理论上该有的样子｜42章经》From WeChat official account: 42 Chapters, author: Qu Kai. Original title: "Optimize for win rate, not odds — do things the way they theoretically should be done"

曲凯：今天我们又请到了albert。之前他在我们这边做过两次播客，是非常受欢迎的嘉宾。Qu Kai: Today we’ve invited Albert again. He’s been a guest on our podcast twice before and is a very popular guest.

从2023年到现在差不多三年，你做了不少尝试，上一款产品也做到小几千万刀ARR，能从头讲讲这三年的思路吗？It’s been about three years since 2023. You’ve made quite a few attempts, and your last product reached an ARR of several tens of millions of dollars. Could you walk us through your thinking during these three years from the beginning?

Albert：2023年录**期播客的时候，我的状态还是看赔率大于看胜率：Albert: When I recorded the first podcast episode in 2023, my mindset still favored odds over win rate.

假设这件事发生了，能创造多大价值、具备多强壁垒？从这个视角出发，你推导出来的结论自然是——怎么用好AI这把大榔头，去创造已经被验证过的商业形态。那因为我自己一直做连接和内容出身，方向就很明确：找一种以内容为核心载体的连接平台，而要定义这样一种产品，核心是要找到一种新的媒介。Assuming something happens, how much value can it create, and how strong a moat can it build? From that perspective, the natural conclusion is — how to use AI as a big hammer to create a business model that has already been validated. Since I’ve always been in connectivity and content, the direction was clear: find a connection platform centered around content, and to define such a product, the core is to discover a new medium.

我们当时押注的是互动内容，它和视频、图像这类消费式内容有本质区别，而AI又解锁了coding能力，让创作这种内容变得更容易。我们前后做了两个demo：一个是偏图像和视频方向的2D交互方式，另一个把交互改成手机上的摇杆控制器，PC端也做了一个更像游戏的可互动空间。At that time, we bet on interactive content—fundamentally different from consumptive content like video and images—and AI unlocked coding capabilities, making it easier to create such content. We made two demos: one was a 2D interaction leaning toward images and videos, the other changed interaction to a joystick controller on mobile, and we also built a more game‑like interactive space on PC.

做完以后，我**的体感是：我很难回答「为什么我不去玩王者荣耀、不去刷抖音？」。After finishing, my biggest feeling was: it was hard for me to answer, “Why wouldn’t I just play Honor of Kings or scroll through Douyin?”

这让我意识到一个规律：在内容市场里，创造门槛越高的模态，供给越稀缺。用户的时间有限，只会消费最头部的1%的内容，而当时AI只能做出60分、70分、甚至80分的内容——但80分的内容对消费端来说就是垃圾。所以for消费、做内容这条线，肯定不work。This made me realize a rule: in the content market, the higher the creation threshold of a modality, the more scarce the supply. Users have limited time and only consume the top 1% of content. At that time, AI could only produce content that scored 60, 70, or even 80 points—but 80‑point content is garbage for consumption. So the line of making content for consumption definitely wouldn’t work.

For表达、做工具可能有价值，但这里有个关键区分：用户的创作动机到底是自我表达，还是更功利的目的——赚钱、获取影响力？模态创作成本越高，供给驱动的特性就越强。如果你的切入点是降低门槛，就会牺牲自由度，变成全家桶——因为它们之间有天然的tradeoff。For expression, making tools might have value, but there’s a key distinction: is the user’s creative motivation self‑expression, or a more utilitarian goal—making money, gaining influence? The higher the cost of modality creation, the stronger the supply‑driven characteristic. If your entry point is lowering the barrier, you sacrifice freedom and become an all‑in‑one package—there’s a natural tradeoff between them.

AI有可能突破这个tradeoff。但即便工具侧有好的解法、技术也到位了，分发侧仍然有很大障碍——拿互动内容来说，如果它更偏游戏形式，你很难做出超越Steam、App Store或TapTap的价值和体验。AI might break this tradeoff. But even if there’s a good solution on the tool side and the technology is in place, distribution still faces big obstacles—taking interactive content as an example, if it leans more toward a game format, it’s hard to create value and experience that surpasses Steam, App Store, or TapTap.

所以我回过头来重新思考，为什么我要做一个连接平台？我后来意识到这种想法受中国市场的影响很深。因为在这里，如果你没有强规模效应、强网络效应的平台形态，就很难在和巨头的竞争中获得优势，胜率会非常低。So I went back and rethought: why did I want to build a connection platform? I later realized this thinking was heavily influenced by the Chinese market. Because here, if you don’t have a platform with strong scale effects and strong network effects, it’s hard to gain an advantage in competition with giants—the win rate would be very low.

但当我去研究美国市场以后，我发现那边其实存在大量的niche market，显然是有一定胜率空间的。But when I studied the US market, I found that there are actually a lot of niche markets there, which obviously have some room for win rate.

所以到2024年初，我完成了一个切换，从赔率driven，变成了胜率driven：去研究哪些技术已经成熟，而哪些真实的用户问题没有被解决。So by early 2024, I made a switch: from odds‑driven to win‑rate‑driven: researching which technologies were mature, and which real user problems remained unsolved.

曲凯：你说2024年之前，你应该基本都在优化赔率。这其实也是绝大多数创始人在做的事情。能不能再多讲讲，你是如何理解这两个问题的？Qu Kai: You said that before 2024, you were basically optimizing for odds. This is actually what most founders do. Can you elaborate more on how you understand these two concepts?

Albert：大多数创业者优化赔率的原因很简单，VC也是在优化赔率。这样对齐，更容易拿钱。Albert: The reason most entrepreneurs optimize for odds is simple: VCs also optimize for odds. Aligning this way makes it easier to get funding.

但上一代真正成功的企业家，其实基本都在优化胜率。只是有些人运气更好，因为他碰到的赛道本身赔率就非常高。比如张一鸣是一个非常保守的人，是一个标准的胜率驱动者。我之前问他，我应该怎么选创业方向？他反问我：为什么不做更有把握的事情？But the truly successful entrepreneurs of the previous generation were basically optimizing for win rate. It’s just that some people had better luck because the track they encountered itself had very high odds. For example, Zhang Yiming is a very conservative person, a standard win‑rate driver. I once asked him how I should choose a startup direction. He asked me back: why not do something you’re more certain about?

张一鸣最有把握的事是信息分发，他在PC互联网的尾巴就在做搜索。出来做字节跳动，**个做的是内涵段子，到2014年，当时海内外已经有很多视频产品了，但他还是say no。一直等到2016年，很多条件都成熟了，才正式下场做视频。What Zhang Yiming was most certain about was information distribution. He was already working on search in the tail of the PC internet era. When he started ByteDance, the first thing he did was Neihan Duanzi. By 2014, there were already many video products at home and abroad, but he still said no. He waited until 2016, when many conditions were ripe, before officially entering the video space.

黄峥也很典型。PC时代做电商，一直在看供给侧的变化、流量端的变化，也一直在供应链里做很多生意。后来拼多多这个结构性机会出现，他就抓住了。Huang Zheng is also typical. In the PC era, he was doing e‑commerce, always observing changes in supply and traffic, and also doing a lot of business in the supply chain. Later, when the structural opportunity of Pinduoduo emerged, he seized it.

王兴做美团，看起来好像和他们之前做的事情完全不一样。我有次还专门问过王慧文这个问题，他的回答是：在那个时间点，所有做团购的人里，懂线下的没有他们懂线上，懂线上的没有他们懂线下（校内网时期涉及大量的线下地推和管理工作）。Wang Xing starting Meituan seemed completely different from what they had done before. I once specifically asked Wang Huiwen this question. His answer was: at that time, among all people doing group buying, those who understood offline didn’t understand online as well as they did, and those who understood online didn’t understand offline as well as they did (during the Xiaonei.com period, they were involved in a lot of offline promotion and management).

所以看起来是突然转向，但他们都是在上个时代积累一些东西，到了这个时代去做。这是一个很典型的优化胜率策略。So it looks like a sudden turn, but they all accumulated something in the previous era and then did it in this era. This is a very typical win‑rate optimization strategy.

真正**的企业家，基本都在优化胜率，没有人在真正优化赔率。优化赔率，本质上就很像赌博。Truly first‑class entrepreneurs are basically optimizing for win rate; no one is really optimizing for odds. Optimizing for odds is essentially like gambling.

曲凯：我原来一直以为优化胜率或赔率只是不同的策略。按你的意思，它反而不是一个选择题，而是一个更接近「对不对」的问题？所以到底什么叫优化赔率？它会不会本身就是个伪概念？Qu Kai: I used to think optimizing for win rate or odds were just different strategies. According to you, it’s not a choice but rather closer to a question of “right or wrong”? So what exactly does optimizing for odds mean? Could it be a pseudo‑concept in itself?

Albert：比如你说「我要做下一个抖音」，这就是优化赔率。因为它足够大，所以我要做它——这本身就是个伪命题。如果你真的是在优化胜率，那你在**天其实说不出来「我要做抖音」。**天你应该说的是：你到底要解决什么问题。Albert: For example, if you say “I want to make the next Douyin,” that’s optimizing for odds. Because it’s big enough, so I want to do it—this itself is a false proposition. If you’re really optimizing for win rate, on the first day you wouldn’t say “I want to make Douyin.” On the first day you should say: what problem do you really want to solve?

曲凯：所以你从优化赔率切到优化胜率，最终在行为上真正发生了什么变化？Qu Kai: So when you switched from optimizing odds to optimizing win rate, what actually changed in your behavior?

Albert：更具体一点说，就是尽量选择变量更少、自己能控制得更多的事情。如果一个事情变量太多、不可预测性太高、超出我能力范围的东西太多，那我就尽量不选。Albert: More specifically, try to choose things with fewer variables and more within my control. If something has too many variables, too much unpredictability, too many things beyond my ability, then I try not to choose it.

曲凯：那你觉得2026年做什么样的事是正确的？Qu Kai: So what do you think is the right thing to do in 2026?

Albert：首先，我觉得AI这件事情还是很长。Albert: First, I think AI is still a long game.

多模态的理解能力，这件事情肯定值得被利用。我一直在说的是视频模型和图像模型，而不太想直接说多模态。因为在我看来，多模态更代表的是理解能力，而不是生成能力。Multimodal understanding capabilities are definitely worth leveraging. What I’ve been talking about are video models and image models, not directly multimodal. Because in my view, multimodal represents understanding ability more than generation ability.

曲凯：那你有没有想过，假设未来技术已经完全成熟，调用成本也足够低了，最酷的产品会是什么？比如就当是在写一篇科幻小说的话。Qu Kai: Have you ever imagined, assuming future technology is fully mature and invocation costs are low enough, what would be the coolest product? Like if you were writing a sci‑fi story?

Albert：我最近其实在想一个很有意思的故事。有一个人，他是AI的信徒，他相信一切都是注定的，一切都是可以被证明的。于是他试图集合全世界的算力，把所有「已经被证明的约束」都注入到一个系统里……Albert: I’ve actually been thinking about a very interesting story recently. There’s a person, a believer in AI, who believes everything is destined and everything can be proven. So he tries to gather the world’s computing power and inject all “proven constraints” into a system…

然后有一天他发现，这个演化世界停在了某一个时刻。因为在那个世界里，也有人开始集合所有算力，去预测自己的未来。于是这个循环又开始了。Then one day he finds that this evolutionary world stops at a certain moment. Because in that world, someone also begins to gather all computing power to predict their own future. So the cycle starts again.

所以最后你会发现，对未来的全部想象，本质上都在预测未来本身。So in the end, you’ll find that all imagination about the future is essentially predicting the future itself.

曲凯：我记得马斯克好像讲过，现实世界本身是虚拟的概率非常高。Qu Kai: I remember Musk once said the probability that the real world itself is virtual is very high.

Albert：对，最后其实就是那个推论。未来的显化，本质上就是不断地预测未来。Albert: Yes, in the end it’s that inference. The manifestation of the future is essentially continuously predicting the future.

42章经

思考事物本质

频道: 商业消费