引子
AI 知识搜索引擎 Perplexity 的创始人 Aravind 在接受采访时说：
The founder of the AI knowledge search engine Perplexity, Aravind, said in an interview:
最好的点子往往是那些你明说出来，人们还会觉得，哦，谁会去实施这个？这听起来真是天方夜谭。
The best ideas are often those that you explicitly state, and people still think, "Oh, who would actually implement this? It sounds like a fairy tale."
为何很多的聪明人看到一个很好的新想法，却会无动于衷？
Why do many intelligent people see a great new idea but remain indifferent?
为何大厂的资金和人员都非常充盈，为何在创新方面总是慢人一步？
Why are the funds and personnel of large companies always abundant, but they are always lagging behind in innovation?
Aravind 的回答是： Aravind's answer is:
只有结合了坚定的决心和深入骨髓的信念，这个想法才有可能成功。
Only with a firm determination and deep-rooted belief can this idea succeed.
本文会展开讨论一下这个问题。 This article will discuss this issue in depth.
创新者的窘境再次上演 The dilemma of the innovator is playing out again.
创新者的窘境 Innovator’s Dilemma 是由哈佛大学商学院克里斯滕森教授提出的管理学概念,描述了一个成功公司在面对市场和技术变革时所遇到的困境。其主要观点包括:
The Innovator's Dilemma is a management concept proposed by Professor Christensen of Harvard Business School, describing the dilemma that a successful company faces when confronted with market and technological changes. Its main points include:
- 成功公司往往过于关注现有客户的需求，错失新兴市场和颠覆性技术带来的机会。
Successful companies often focus too much on the needs of existing customers, missing out on opportunities brought by emerging markets and disruptive technologies. - 管理者基于现有商业模式做出的合理决策，反而阻碍了公司拥抱变革，导致最终被颠覆。
Decisions made by managers based on existing business models actually hinder companies from embracing change, leading to eventual disruption. - 决策者面临资源配置和价值网困境，难以平衡当前业务和未来布局。
Decision-makers face the dilemma of resource allocation and value network, struggling to balance current business and future layout.
一个典型的例子是 AI 云厂商。 A typical example is AI cloud vendors.
在2022年，中国公有云厂商的市场份额，阿里云、华为云、天翼云稳居前三，腾讯云由第二位降至第四。而火山引擎则在 Others 里。
In 2022, the market share of Chinese public cloud vendors, Alibaba Cloud, Huawei Cloud, and China Telecom Cloud, remained in the top three, while Tencent Cloud dropped from second to fourth. And Volcano Engine is in Others.
当时腾讯云和阿里云的显卡冗余，没有客户，实在租不出去了，就把英伟达的显卡订单都退了
At that time, Tencent Cloud and Alibaba Cloud had redundant graphics cards, no customers, and couldn't rent them out, so they returned all the NVIDIA graphics card orders.
。
在当时看，这是一个非常合理的决策。 At the time, this was a very reasonable decision.
只有 Others 里不配拥有姓名的火山云还在坚持买显卡，一不小心就买成了国内显卡最多的公司。
Only the volcano cloud in Others that is not worthy of a name is still insisting on buying graphics cards, and accidentally became the company with the most domestic graphics cards.
2022 年11月，ChatGPT 发布了。 In November 2022, ChatGPT was released.
大模型厂商一个一个地成立，都需要显卡训练模型，在阿里和腾讯都没回过神的时候，大家发现卡最多，技术最好的是火山云，都纷纷去找火山云合作。
Large model manufacturers are established one by one, all of them need graphics cards to train models. When Alibaba and Tencent were still not awake, everyone found that the most cards and the best technology were in Volcano Cloud, and they all went to cooperate with Volcano Cloud.
到2023年4月的时候，火山引擎的总裁谭待公开表示，国内大模型领域，七成以上已是火山引擎客户。
By April 2023, Tan Dai, the president of Volcano Engine, publicly stated that more than 70% of the domestic large model field are customers of Volcano Engine.
未来已来，视而不见 The future is already here, but many people choose to turn a blind eye.
为什么在2022年的时候，火山云能够这么有远见地囤显卡呢？是因为字节跳动很有前瞻性吗？
Why is Volcano Cloud so visionary in hoarding graphics cards in 2022? Is it because ByteDance is very forward-thinking?
很可能不是的。在地球对面，还有一家公司和火山云一样靠囤显卡获得了竞争优势。
It's very likely not. On the other side of the Earth, there is another company that, like Volcano Cloud, has gained a competitive advantage by hoarding graphics cards.
前几天 Meta 的扎克伯格接收访谈也被问到了这个问题。
A few days ago, Meta's Mark Zuckerberg was also asked this question in an interview.
主持人：你们是怎么顶住股价下跌的压力大笔买入gpu的呢？
Host: How did you withstand the pressure of falling stock prices and buy a large number of GPUs?
扎总：当时我们正在做Reels （类似tiktok），需要大量gpu做推荐算法，也没想到大模型来了。
Zha (CEO): At that time, we were working on Reels (similar to TikTok), and we needed a large number of GPUs for recommendation algorithms. We didn't expect the arrival of large models either.
很坦诚。但是火山引擎在当时囤显卡的理由大概率也是类似。
Very frank. But the likely reason for stockpiling graphics cards by Volcano Engine at that time was probably similar.
为什么这么说？ Why do you say that?
因为字节跳动的大模型在内部启动非常晚。 Because ByteDance's large models started very late internally.
这就涉及到两个非常有趣的故事： This involves two very interesting stories:
一个是字节的战略部门在2022年的所有战略报告里都没有提到生成式AI的内容。属于一个重要的Miss。
One important miss is that the strategic department of ByteDance did not mention the content of generative AI in all strategic reports in 2022.
一个是在2023年初，字节的算法大佬提出要做大模型的时候，竟然被公司拒绝了，于是他离职去了大模型公司创业。
Another story is that in early 2023, when ByteDance's algorithm expert proposed to work on large models, he was actually rejected by the company, so he resigned and started a business at a large model company.
这些故事都是坊间传闻，可能并不是真的，但是实际的情况是字节的大模型确实晚了一步。
These stories are all rumors, may not be true, but the actual situation is that ByteDance's large models did indeed lag behind.
晚了这一步，导致云雀大模型在现在的竞争格局里非常尴尬，coze在中国搞开发者活动还要接入moonshot。
Being late for this step, the lark large model is very embarrassed in the current competitive situation, and coze has to access moonshot for developer activities in China.
一般来说，大公司的战略部门应该能洞悉科技前沿，大公司的决策层也应该高瞻远瞩。
Generally speaking, the strategic department of a large company should be able to insight into the forefront of technology, and the decision-making level of a large company should also have a long-term vision.
但实际上刚好相反。原因其实非常简单： But in fact, it is just the opposite. The reason is actually very simple:
当下的商业太成功了 The current business is too successful
穷则思变，富则不见。 Poverty leads to change, wealth leads to invisibility
被低估的创新 Innovation that is underestimated
分析完了，道理大家都懂，有办法破局吗？ The analysis is done, everyone understands the reasoning, is there a way to break the deadlock?
答案是很难。 The answer is very difficult.
再举一些简单的例子 Give a few more simple examples.
- 某社交网络公司，做了一款AI社交产品，但是数据表现不如原来的产品，上线后两个月就停止了项目。
A social networking company developed an AI social product, but the data performance was not as good as the original product, and the project was stopped two months after launch. - 某知识问答公司，做了一款AI问答产品，但是上线后发现AI问答影响了传统问答的广告收入，上线后很快就在公司内部的利益battle中失败，下线了AI问答功能。
A knowledge Q&A company developed an AI Q&A product, but after launch, it was found that AI Q&A affected the advertising revenue of traditional Q&A. It quickly failed in the internal battle of interests within the company and the AI Q&A feature was taken offline. - 国内两个头部的 AI super ap产品，做了一年数据不如公司现有的非AI产品，在年底绩效评完后，都被替换到了其他部门。
Two top AI super app products in China did not perform as well as the company's existing non-AI products after a year of development. After the year-end performance evaluation, they were all reassigned to other departments.
例子还有很多，其中的 Pattern 很明显。 There are many examples, among which the pattern is quite obvious.
大公司的做事方式，是以年或者半年为单位的绩效考核，做不好就换人。
The way big companies do things is performance evaluation on an annual or semi-annual basis. If you don't perform well, you get replaced.
在一个技术浪潮的初期，技术的能力还有限，GPT4一年了还没到GPT5呢，项目的技术负责人都换了一轮了。
In the early stage of a technological wave, the technological capabilities are still limited. GPT-4 has only been around for a year and GPT-5 hasn't arrived yet. The project's technical leads have all been replaced in a round.
让我们再回味一下 Aravind 的话： Let's revisit Aravind's words:
只有结合了坚定的决心和深入骨髓的信念，这个想法才有可能成功。
Only with a firm determination and deep-rooted belief can this idea succeed.
然而这个决心和信念，太稀缺了。 However, this determination and belief are too scarce.
在4月聊的接近50个人里，有决心和信念的不超过5个。
Among nearly 50 people I talked to in April, no more than 5 had determination and conviction.
有的人迷茫，在技术的新浪潮里，很正常。 Some people feel lost, which is quite normal in the tide of technology.
有的人等待，做好准备，等待copy Some people wait, prepare, and wait to copy.
有的人摆烂，活得太好了，copy paste 都懒得搞一下。
Some people are pretending to be bad, living too well, and can't even bother to copy and paste.
只有很少的人在创新。 Only a few people are innovative.
因为创新的成本被低估了，不仅是需要钱、需要人、还需要决心和信念。
Because the cost of innovation is underestimated, it not only requires money and people, but also determination and belief.
但是创新的价值也被低估了。 But the value of innovation is also underestimated.
想起当年的招聘市场，早已被瓜分完毕， Boss 直聘借着移动互联网的浪潮，创造了「IM招聘」的范式（和老板聊则只是个噱头，大部分都还是HR），后来被各家竞品抄袭，但是无法阻碍 Boss 的增长。
Thinking back to the recruitment market of the past, which had long been divided up, Boss Zhipin created the paradigm of "IM recruitment" with the wave of mobile internet (chatting with the boss is just a gimmick, most of them are still HR), later copied by various competitors, but unable to stop Boss's growth.
几年前发现，Boss 的月活用户在每年的春节后几乎都会 double，年年如此，就知道它一定会成为第一。
A few years ago, it was discovered that Boss's monthly active users almost double after the Spring Festival every year, year after year, knowing that it will definitely become the first.
前几天看到 Boss直聘的最新数据，已然是行业第一了。
A few days ago, I saw the latest data from Boss Zhipin, which has already become the industry leader.
2023 年国招聘平台 Boss 直聘的营收增长到将近 60 亿元，第四季度的月活跃用户数也达到了 4100 多万。根据数据机构 Questmobile 的报告，这一数字几乎是智联招聘、前程无忧和猎聘三家招牌平台的月活用户数之和。
In 2023, the revenue of the national recruitment platform Boss Zhipin increased to nearly 6 billion yuan, and the monthly active users in the fourth quarter also reached over 41 million. According to the report from data agency Questmobile, this number is almost the sum of the monthly active users of Zhaopin, 51job, and Liepin, the three flagship platforms.
为什么每年的春节后几乎都会 double 呢？ Why does it almost double every year after the Spring Festival?
背后的原因，请关注橘子的汽水铺，我们以后再聊。 The reason behind it, please pay attention to the orange soda shop, we'll talk later.
永久链接：https://quail.ink/orange/p/ai-innovators-dilemma-and-opportunity
Permanent link: https://quail.ink/orange/p/ai-innovators-dilemma-and-opportunity
评论