Communication group| Enter "Sensor Group/Skateboard Chassis Group/Car Basic Software Group/Domain Controller Group", please scan the QR code at the end of the article , add Jiuzhang Assistant , be sure to note the name of the exchange group + real name + company + position (no remarks Unable to pass friend verification)
This article is the last article in the series of autopilot essays . Click on the following titles to jump to the first three articles in this series :
◆ What is the difference between "Remove HD Map" and "Light HD Map"? What is the challenge of landing? | One of nine chapters on autonomous driving
Editor | Su Qingtao
A big crux of the failure of the autopilot business of OEMs: people who do technology don’t understand the market, and people who do market don’t understand technology; they also think that they are “high-level and don’t need to know too many details.”
Strictly speaking, this cannot be regarded as a serious "article", but a "patchwork" of some fragmented observations and thinking.
The main topics covered in this article are:
1. "Full-stack self-research" is the small-scale peasant economy
2. Professional managers: self-study can prove that "I have value"
3. The collapse of "full-stack self-research" and its cause analysis
4. OEMs began to "return to the essence of business"
5. Requiring suppliers to "deliver in a white box" is to dig a hole for themselves
6. An underestimated problem: the "reading comprehension ability" of the bosses of the main engine factory is not good
1. "Full-stack self-research" is the small-scale peasant economy
In the past two years, the author has heard countless times on various occasions that various suppliers have painstakingly analyzed for OEMs that self-developed certain technologies are "not cost-effective"——
The perception algorithm should be a standard product and does not need to be differentiated. Since it is a standard product, it is more cost-effective to buy from a supplier than self-developed (in terms of cost and performance).
The algorithm architecture is evolving rapidly. You may burn billions to achieve some good results. Then, after a few years, new and better algorithms come out. Wouldn't the money just burn in vain?
Self-developed, quality control may not be better than cooperation with suppliers. When you buy something from a supplier, you can "destroy him to death", but among internal colleagues, in many cases, even if the other party is doing poorly, you have to show some sympathy, but this will lead to constant changes in your standards. reduce.
Chips, even if the self-developed chips are successful, if the monthly sales volume of the car is only a few thousand, 10,000 to 20,000, even if each car is equipped with 2 self-developed chips, the demand is not much, who is willing to OEM for you? ? Even if there is a fab willing to manufacture, the price is not low, and the delivery date will not be prioritized.
The self-developed "yan" means "grinding" in Chinese characters, which means that you need to grind every particle into powder. will be super high.
In case the self-research fails, you are too embarrassed to go back to the supplier, because it is embarrassing to say it. What's worse, after the failure of self-research, you will find that it is difficult for you to find a reliable and hardcore supplier for a while, because, during the short period of your self-research, the supplier has already focused on other customers or the business that was previously customized for you has been cut off.
But the fathers of the main engine factory seem not to listen to it, and still go their own way. Does this mean that these OEMs are all "strong-minded and not easy to be brainwashed"? No, judging from the results, they are called "pursuing the highest cost and lowest profit at all costs".
One day, the author suddenly realized that the income of farmers is lower than that of people working in cities, which is largely caused by "self-controllable" and "full-stack self-research"!
In cities, most people do their own jobs well, and those that they don’t know how to do and that are not cost-effective are all bought in the market; while in rural areas, almost every family does it within their own ability. "Vertical integration", the most common cases are:
The same farmer’s family planted grains such as wheat, buckwheat, corn, and millet at the same time, and vegetables such as peppers, tomatoes, flower melons, eggplants, beans, green vegetables, cabbage, Chinese cabbage, and potatoes, and planted apple trees, pear trees, Plum trees, apricot trees, peach trees and other fruit trees raised a pig, a cow, a few chickens, a few rabbits, a few sheep, and some families even planted medicinal materials at the same time... and , Some farmers are also carpenters, construction craftsmen, and even drive tractors to pull goods for others.
Is it really necessary to be so big and comprehensive? Every industry/product has not enjoyed the scale effect, the cost is extremely high, and the profit is extremely low! Wouldn't it be better if all the resources (land, capital, and energy) were used to do one or two things that you are best at or "the most profitable", and buy the rest in the market? It is true that there will be periodic market failures in this way, resulting in oversupply of certain types of products and short supply of certain types of products, but after several rounds of dynamic adjustments, a balanced state will always be reached, and in the end, everyone will be the beneficiaries.
But judging from the author’s growth experience, farmers don’t accept the division of labor very much—it’s obvious that I can raise chickens and lay eggs myself, so how can I be willing to spend money to buy eggs from other people’s homes? The vast majority of farmers do not have the concept of opportunity cost, let alone the awareness of "improving professionalism through focus and improving production efficiency through professionalism".
Marxists often say that "the peasant class is not a representative of advanced productive forces", but in my opinion, many large companies would rather make the cost high and the efficiency low, but also insist on "full-stack In terms of business logic, there is no essential difference from the practice of the peasant class that "does not meet the development requirements of advanced productive forces".
The author has communicated with people in the automatic driving department of several new forces. Many of them think that the "full-stack self-research" emphasized by their boss is either just a slogan or a joke.
Under the current competitive situation, for OEMs, is the most important thing to pursue the soul, or to keep the body first? The stock prices and sales of several major new forces have fully explained the problem.
Some time ago, a friend from the automatic driving department of an OEM wrote in "From Stockpiling to Radical Reduction, Why Has Automatic Driving Changed?" "It is mentioned in the article:
It is true that you can save your soul if you choose to work with teachers and mobilize the whole stack for self-research, but if you are a foolish child conceived by painstaking efforts, you will lose everything and there will be no day to turn around.
And choose to continue the old model of "parts-function-system" all outsourcing, even if it will be labeled as "foundry" and "comprador", but the advantage is that it can be quickly launched, mass-produced and sold, and realized profit, and more importantly Yes, you can share costs and risks with downstream suppliers, which is a more practical choice after putting aside the rhetoric at this stage.
During the auto show, the author chatted with the CEO of an autopilot company about the topic of self-developed autopilot by the OEM. At that time, the CEO said: "Tesla's real money-making link is not autopilot, but intelligent manufacturing. Many people I didn't get this point."
2. Professional managers: self-study can prove that "I have value"
Since self-research is not cost-effective, and it is difficult to succeed, why do you still need to conduct self-research?
Recently, a friend told a story: In a new head force, once the boss asked the R&D department to make something, and the person in charge of the R&D said which supplier did a better job and could "buy it directly for use". Then, he was scolded by the CEO: "Then what do I want you to do?"
In more cases, whether or not to "full-stack self-research" is not just a question of cognition, but also a question of redistribution of benefits.
Even if the boss has realized that "full-stack self-research" is not the optimal solution, where can he let the people in relevant internal departments go? In order to convince the boss that their existence is valuable and to obtain more resources, those departments will definitely emphasize the necessity of "full-stack self-research", even if "internal customers" already think that this "internal supplier" In fact, far less than external suppliers.
As a result, the competitiveness of enterprises has been weakened step by step.
3. The collapse of "full-stack self-research" and its cause analysis
1.
OEMs no longer hold data tightly in their hands
In 2018, when the author talked with the CEO of a self-driving company about the issue of data ownership, the CEO said: "Although the data belongs to the OEM, in fact, based on the level of most OEMs, even if the data is given to them, They couldn’t use it either.”
At that time, the author conceived the issue of "decoupling" data ownership and usage rights, and asked people intensively in 2020, but when asked about this issue, everyone was vague. However, from last year to this year, the answer to this question has become more and more clear: the ownership of the data still belongs to the OEM, but the algorithm company that makes the solution for the OEM has the right to use it for free.
Algorithm companies can't get all the data, they can only get the corner case data that OEMs find themselves unable to handle after encountering difficulties. It seems that there is not much data, but this part is precisely the most valuable data—it is equivalent to that the OEM has already helped the algorithm company to do a round of data screening.
A friend joked: "The main engine factory has spent a lot of energy doing the work of 1+1=2, and handed over the work of solving the equation to the algorithm supplier."
2.
Self-developed algorithm failed
In the past month or so, for those in the autonomous driving industry, the most shocking thing was that many OEMs were exposed that the bubble of "full-stack self-development" began to burst.
For example, most of the executives of Caraid, a software company owned by Volkswagen, have been laid off. For example, the self-development of autonomous driving by a domestic OEM was exposed by the media as "a power game that wasted ten years".
People without "software thinking" can't manage the autonomous driving team well
After the news of Caraid’s layoffs came out, an industry veteran who had worked in OEMs, custom driving companies, and chip manufacturers commented on WeChat Moments:
Diess, the former CEO of Volkswagen, realized that "software-defined cars" is the general trend, but Volkswagen itself is still hardware DNA. After driving Diss away, it was only natural to take a knife on his Cariad.
In fact, even if the reform operator is not kicked out due to conflicts of interest, it will be very difficult for a traditional OEM like Volkswagen to succeed in self-development of autonomous driving.
The most critical reason is that in companies with hardware NDA, the management who have been engaged in machinery for a lifetime generally lack "software thinking".
Tesla developed Autopilot 2.0 when the autopilot team had less than 200 people, while the total number of Volkswagen Cariads is several thousand, and the number of people who specialize in autopilot is estimated to be one or two thousand. Why does Volkswagen need so many people? ? In the eyes of people with software thinking, in many cases, the value created by a powerful engineer exceeds the value created by thousands of ordinary engineers; while people with only hardware thinking believe that "more people are more powerful."
Three years ago, when the author communicated with the CEO of a star autonomous driving company, the other party said: "If I see a company saying that it will recruit tens of thousands of people to do autonomous driving, I will conclude that 'they must not succeed'." Yes, the way of thinking How can it be successful if it is out of step with the requirements of the times?
In addition, it is a routine operation in traditional OEMs to let people with mechanical background lead software talents. Software talents find that their leader is a "layman", and the requirements raised and management methods are all absurd. How can they be willing to accept it?
Automobile Business Review's article on May 8 "Anxiety of the public, CARIAD pays the bill? " mentioned that a software engineer working for the Volkswagen brand said that for a long time, many of Volkswagen's decision-makers were born mechanically, and most of the code was outsourced, which led to the lack of code writing skills for Volkswagen internal managers. talent and management. The current situation of CARIAD may be caused by the fact that the highest level does not fully understand the laws of the software.
Moreover, Chapter Nine mentioned in the previous article that many leaders who lack software thinking believe that "software is intangible, invisible, and worthless, and the software of suppliers is only suitable for prostitutes." What about respecting the software talents of your company in your heart? This is a big suspense.
An expert from an autonomous driving giant analyzed: "The full-stack self-research of the main engine factory is not to say that it can't be done. Even if it can be done, it will require a large number of layoffs, leaving only a few people for maintenance. Because people are too expensive. Why not cut it. Engineers are well aware of this prospect." Therefore, engineers do not have enough motivation to pursue self-development to "succeed as soon as possible."
3.
Self-developed chip failed
Leapmotor is the earliest self-developed self-driving chip among the domestic OEMs, and Dahua, the major shareholder of Leapmotor and the partner of the self-developed chip, has rich experience in semiconductors. It is said that the self-developed chip of Leapmotor has the possibility of success, but now , this matter has no more context.
Several leading new forces have performed very high-profile on self-developed self-driving chips, but how is the progress, and when will the mass production be launched?
According to the information recently learned by Jiuzhang Zhijia, they all plan to launch a new central computing architecture around 2025, and this time point coincides with the launch time of the next self-driving chip of an international chip manufacturer. There are not small differences in the models of several OEMs, but they were able to set the launch time of the new architecture platform in the same year, which only shows that the launch time of these new architectures is determined by the launch time of the latest products of the chip supplier. of.
The launch time of the central computing architecture follows the launch time of the supplier's new products, which further shows that their model department did not put their hopes on their own self-developed chips at all.
There is another fact that can be proved: a key member of a self-developed chip team of a new force said when communicating with Jiuzhang Zhijia, "I am worried that I will be laid off soon." "The boss also found that self-developed chips are indeed too difficult. With our current team capabilities, it is unlikely to produce car-grade chips."
Today, this new force has lowered its expectations for self-developed chips: it does not insist on producing usable chips, but only wants to quickly grasp some of the underlying development logic of the chip, so that it will know how to do chip selection later. You can also have a greater say if you raise your needs.
After the boss recognizes the reality and gives up fantasy, the ineffective investment will be reduced, and the company may live a healthier life.
Having said that, it is inevitable for OEMs to develop their own self-driving chips, failure is inevitable, and success is only accidental.
First of all, the level of chip leaders recruited by OEMs is completely incomparable with the founders of independent chip companies.
For example, the person in charge of chips recruited by a certain main engine factory used to make consumer-grade chips and scrapped several generations of products in succession. Even consumer-grade chips can't do well, so why can we do well with car-grade chips?
Just think about it, if these chip leaders are really high-level and have developed an autopilot chip by themselves, why doesn't he raise money and start a chip company by himself, instead of receiving that salary at the OEM?
Secondly, the learning speed of the head of the chip of the main engine factory is not as fast as that of the founder of the independent chip company.
This is easy to understand. The business negotiation between the company and the company pays attention to "level equivalence". The person in charge of the chip of the main engine factory is a professional manager, and most of the external personnel who usually have in-depth contact with him are occupations "equal to his level" Managers; and the founders of chip companies are usually CEOs, and most of the outsiders who usually have in-depth contact with him are CEOs of other companies.
The process of dealing with another person is also a process of "collecting data" from others. These data can be used to iterate our own algorithms (cognition); even, in many cases, we can directly obtain other people's algorithms . Obviously, the quality of data collected and the quality of algorithms collected by those who have been dealing with managers for a long time are not as good as those who have been dealing with CEOs for a long time. In the long run, the gap between the former's algorithmic capabilities and the latter's will become wider and wider.
On the other hand, the magnitude of the decision is also different. In general, the decisions that professional managers of OEMs usually make are much less difficult than those made by CEOs of external chip companies, which also affects the iteration speed of his "decision-making algorithm".
Also, the chip leaders of OEMs are professional managers. Can the sense of mission of these professional managers be compared with the founders of independent chip companies?
Many bosses who are pure businessmen do not really believe in the "sense of mission". They think that as long as they are willing to spend money to recruit a good person, everything can be done. However, if you don't have enough sense of mission and don't love the company like you love your own children, then it's not appropriate to be number one.
When the corporate culture is strong enough, professional managers may also have a strong sense of mission. However, how many OEMs dare to say that their corporate culture is stronger than that of Huawei?
Finally, the organizational culture of OEMs makes it difficult to retain excellent chip talents.
Except for a few OEMs whose corporate culture is a bit "technical", most OEMs are essentially manufacturing cultures (after all, another name for OEMs is "automobile manufacturers"). ", "Technology companies", but many core executives come from traditional car companies. These executives may "don't know how to socialize with people without drinking", which is essentially "manufacturing culture".
How can such a culture attract and retain excellent chip talents?
Just imagine, if such a cultural and organizational culture can guarantee the success of self-developed chips, why doesn't the person in charge of self-driving cars stay to make chips, but join an outside chip start-up company?
Some time ago, after OPPO decided to abolish the entire chip team of Zeku, I was deeply impressed by these two comments.
One is a comment from Vernacular IC:
As soon as his head got hot, he would frantically spend money and poach corners everywhere, making job-hopping a common practice in the industry. In the end, all of them were evacuated, leaving a pile of chicken feathers. Tens of billions of dollars have been spent, which is not helpful to the development of the industry, and it can even be said to be a disservice.
The second is a review article from "Xinmou" "Fear the Law of Industry, Respect the Mode of Division of Labor and Cooperation", which contains the following paragraphs:
Each enterprise can only do what it is good at. Counting the most successful giants in the chip industry, they rarely dare to cross the border. Companies like Samsung and Huawei that do both mobile phones and chips have an inevitable reason for their success. First of all, when Huawei and Samsung made mobile phone chips, they already had a deep accumulation in radio frequency, power, analog, and communication chips. They made mobile phone chips by following the trend. In terms of talents, when Huawei manufactures mobile phone chips, it can form a combat force by integrating different internal departments. Unlike OPPO’s core-making, no matter technology accumulation or R&D team building, we must start from a blank.
Creating chips and market acceptance of chips are two logics. Even if OPPO succeeds in making cores, for a long time, OPPO mobile phones dare not use their own chips. The semiconductor dictatorship is not satisfied, how many people with strong strength and high hearts, regardless of the laws of the industry, devote themselves to making cores, but in the end they fall into the sand.
If you like seafood, you book a beach; if you like food, because there are too many people and you have to queue, you buy the whole restaurant in a fit of anger. It’s fine if you are a real local tyrant, but if you borrow money to buy a restaurant, you are often the first to starve to death in the cold winter of the industry.
If these few paragraphs are used to evaluate automakers' self-developed self-driving chips, they are also particularly appropriate.
4.
The voice of a former employee of an OEM
An autopilot engineer/project manager of a joint venture car company recently wrote in an article titled "An Atypical Story: A Car Company's Full-Stack Self-Research, I Graduated from a Full-Stack":
The self-developed project, which was ambitious at the beginning, is now facing tremendous pressure, the research and development progress is very slow, and there are more and more voices of doubt. "Others have already achieved functional mass production, so is it meaningful for us to do our own research?"
There also doesn't appear to be any plan above for what to do next. On the one hand, I saw the rapid development of the industry, and peers were arguing fiercely around various technical routes and business models. On the other hand, I saw the depression within the department, and the lack of a goal that could motivate everyone to unite.
From a technical point of view, the development work is not very difficult, but what really hinders the progress is not only the dependence on the delivery of external suppliers, but also the internal power of gradual forking. It's like in a big group photo, everyone is facing in different directions, some don't care, some are maverick, and some are at a loss.
I became more and more silent, and I didn't want to talk, let alone say nice words. Day by day, my resistance to some things became more and more obvious. I persisted in vain, like a joke, but not funny at all.
I can't convince myself, and I don't want others to waste their precious time with me. So, I ran away.
4. OEMs began to "return to the essence of business"
Of course, after several years of hard exploration, more and more OEMs have begun to recognize the boundaries of their own capabilities, so they no longer talk about "full-stack self-development".
During the auto show, a person from the automatic driving department of a leading new force said at a salon: The three leading companies have developed a lot of self-developed products. Now it is a differentiated selling point, and it may be a disadvantage of driving up costs in the future.
At present, more and more OEMs no longer pursue full-stack development, but hope to master full-stack capabilities, that is, "full-stack controllable". The core goal is to ensure the integrity and security of the supply chain, not to replace supply business.
Also during the auto show, the vice president of a research institute of an OEM said: The closer to the user needs and the closer to the differentiated part, the more it should be developed by the downstream link close to the user; the closer to the commonality, the closer to the underlying part , the more it should be jointly built by the industry.
According to the news from Western Securities on April 21, the autonomous driving domain controller of a leading new car maker will be provided by a Tier 1 (listed company). What Jiuzhang learned from the industry is that this Tier 1 company can also provide application layer algorithms.
According to the conclusions drawn by Jiuzhang Zhijia in 2022, generally, for high-end models + large computing power platforms, OEMs hope to develop their own perception algorithms, while for mid-to-low-end models + medium and low computing power platforms, OEMs will Hope to use the supplier's software and hardware integration solution. From this, it is speculated that the algorithm of the low-end models of this new force is also most likely provided by this Tier 1 company.
5. Requiring suppliers to "deliver in a white box" is to dig a hole for themselves
In order to reduce the dependence on suppliers, many OEMs will force suppliers to provide "white box" solutions. In order to get orders from OEMs, most suppliers will choose to compromise. However, suppliers who are required to deliver in a white box are not fools. In order to protect their long-term interests, most of them will "keep a hand".
The hands left by the suppliers are equivalent to a pit in the psychological experience level of the OEM.
1.
White box delivery in the eyes of autopilot product managers and project managers of OEMs
The products delivered by the white box are generally outdated, and the scenarios that this version of the algorithm can cover are far from the latest algorithms. Are you asking me to update the version of the algorithm? This is what I have iterated in practice with other customers. This property right does not belong to me, so I cannot give it to you; but if you are willing to add a lot of money, we can discuss it.
Only upper-level algorithm modules are provided, but necessary interactions involving middleware may not be provided;
The real core and underlying logic may not be provided;
There are only algorithm codes without detailed notes, so people from OEMs may not be able to fully understand them;
The software adaptability is limited. If the product is upgraded and the computing platform is changed, such as Nvidia is replaced by Huawei, then the software algorithm module needs to be updated. At this time, the supplier may not participate again;
If it is a partial white box delivery instead of a full set of delivery, then the matching and coordination among multiple suppliers will also be a risk point;
White box delivery, the code cannot run after getting it, and I want to develop it myself but can't find a starting point. There is no documentation for the relevant development process.
2.
White box delivery in the eyes of autonomous driving product managers and sales directors of supplier companies
They can't get the AI model, the code has been compressed;
Usually, the electronic and electrical engineers of OEMs are familiar with the application layer, but not very familiar with the underlying configuration and interface definition of the middle layer. Therefore, even if we deliver a white box, when the model is changed, the underlying code and parameters They may still not know how to modify and deploy;
For hardware and software delivered in a white box, not only is the price quoted for the first delivery much higher than that delivered in a black box, but also a “hour fee” is charged for each iteration of the version.
As for the "working hours fee" charged when the version is iterated, didn't the two parties agree in advance, and the conflict broke out when an upgrade was needed?
Faced with such a question, the salesman of the supplier replied: "There is no problem with the agreement. I just need to write on the quotation, 'Quotation based on current demand, and xx has the right to update the quotation' if the subsequent demand changes. You propose When you discover new requirements, or when you find that the code you currently deliver is not enough to support your needs, it is the beginning of the supplier’s collection of money.”
The previous black box delivery model can guarantee long-term supply, and these so-called "working hours" can be allocated to the mass production cost and absorbed by the supplier internally; while the white box loses the guarantee of mass supply, so the foundation of trust It is lost. In this case, of course, the supplier can make a fortune and continue to dig holes for the OEM.
Even, it is obvious that you can iterate your requirements at one time, but I pretend not to know, and I don’t tell you, so that you continue to find that there are still many functions that need to be added—the white box is delivered to you with version 1.0, and you will eventually mass-produce it. It is also possible to use version 20.0.
There are also more ruthless ones, "hiding" the timing in that code, so that in the first period of time, the code can run normally, but after a while there will definitely be problems.
Suppliers are definitely not willing to do one-time transactions, so they will find ways to make OEMs pay them continuously. Doing this will not only increase performance, but they can also tell other customers: "XX OEM is in long-term cooperation with us."
Look, OEMs thought that white box delivery was a one-shot deal, but the supplier turned it into a "long-term cooperation".
A product manager of a supplier company said triumphantly: "I think OEMs that require suppliers to deliver in a white box are stupid."
The CEO of a self-driving company said that the white box is actually just a placebo and anesthetic given to OEMs by suppliers, which cannot improve the OEM's self-research capabilities at all.
China used to play "market for technology" with Germany and Japan. As a result, the market was lost. Have you learned the core technology? Take all those books from the library home, are you a writer? Buy a set of "Historical Records" and come back, can you become Sima Qian tomorrow?
Taking the control software in the engine industry as an example, there are countless engineers who have left Delphi, but can anyone replicate a Delphi? In my opinion, people are much more important than white boxes. If no one in your team can understand what the white box code means, if you give the code to you, you won't be able to make much trouble.
An executive of an autonomous driving solution company said that considering that the white box solution actually delivers outdated technology, and the price is much higher than that of the black box solution, then, from the supplier's point of view, in many cases, the white box delivery In fact, it is a "more cost-effective" solution than black box delivery.
However, "higher cost performance" in the eyes of suppliers is an IQ tax in the eyes of OEMs who are unwilling to be leeks. "A certain Tier 1 once planned to sell the engine control software white box solution to a domestic OEM, but the OEM said: 'No one can handle me.' They are very smart and unwilling to hand over the white box solution IQ tax."
Listening to it, suppliers look at those OEMs that force them to deliver white boxes, not only has the taste of "you are unkind, don't blame me for being unrighteous", but also seems to be a group of people with high self-esteem who are "reducing dimensionality" on others at the intellectual level strike".
6. An underestimated problem: the "reading comprehension ability" of the bosses of the main engine factory is not good
Many OEMs have poor strategies related to autonomous driving. One very important reason is that those who do technology do not understand the market, and those who do market do not understand technology. If the perception is not done well, how can the quality of decision-making be guaranteed?
How did you come to the conclusion that "people who do technology understand the market, but people who do market don't understand technology"? When the author sorted out the user portraits of Jiuzhang, he found that the key decision makers of OEMs did not read the articles of Jiuzhang Zhijia (hereinafter referred to as "Jizhang"), or even heard of Jiuzhang at all.
Of course, people who know little about Jiuzhang and have insufficient awareness of the value of Jiuzhang will not agree with the logic of the author, but if you do not agree, it will not affect its correctness.
During the Shanghai Auto Show, the author discovered an amazing pattern after in-depth exchanges with many self-driving start-up companies: Overall, in the horizontal direction, the product sales and valuation of a self-driving start-up company, compared with their top management , Middle-level familiarity with Jiuzhang is positively correlated; vertically, a person's rank in a self-driving start-up company is positively correlated with his familiarity with Jiuzhang.
Afterwards, after analyzing some of the backstage data of Jiuzhang, the author found that: It seems that the decision-making quality of the new car-making forces on matters related to autonomous driving is also positively correlated with the familiarity of the founder and the person in charge of Jiuzhang with Jiuzhang. .
(Here we are talking about "positive correlation", not "causal relationship". As for the logical relationship between the two, we will do a more detailed analysis later in this article.)
Next, the author will sort out the logical link between "not reading the nine chapters" and "the quality of decision-making cannot be guaranteed"——
1.
Those who have little contact with Jiuzhang are not considered "true industry insiders"
In February 2022, an autonomous driving product manager of Xiaopeng Motors left a message to the author saying: "In October last year, when I recommended the Jiuzhang Official Account to a friend, I realized that I had more than 100 friends following me." I've written nine chapters, and my friends say I'm a 'true insider' because of it."
At that time, the author joked: "Your friend means that those who haven't followed Jiuzhang, or there are not many people who follow Jiuzhang among friends, are they 'fake industry insiders' or 'haven't entered the industry'?" "That's true," said the product manager.
During the Shanghai Auto Show, a colleague of the author participated in a media group interview organized by a self-driving start-up company, and said: "Their employees basically know Jiuzhang, but the CEO has never heard of it." The author said arbitrarily. : "Then the CEO must be new. Not only has he just joined the company, but he has just entered the industry." A colleague said: "Yes. One of the main purposes of this press conference is to arrange for the newly appointed CEO to appear in front of the public. "
2.
Jiuzhang's user portrait and positioning
Many times, when friends in the industry asked the author "who is the target user of Jiuzhang", the author's answer is:
Investors, brokerage analysts, and senior and middle managers of autonomous driving companies are the most active; followed by technical experts, product managers, and product planning department personnel of OEMs, but excluding middle and senior managers of OEMs (individuals) Except for new forces).
Why are investors and brokerage analysts the most active users?
Because investors are the group of people who are most aware of "knowledge is power".
The reading volume of the nine chapters is even strongly positively correlated with the capital market’s attention to the custom driving industry and specific tracks. Two years ago, a new colleague of the author spent more than a month writing an article on the theme of simulation. The reading volume was very average. He was very frustrated. Capital has not yet paid attention to this track.
Under the guidance of the core value of "non-necessary, no nonsense", the biggest highlight of the nine chapters article is "high information density, low nonsense rate", which helps investors who have the appeal of "understanding an industry in 7 days" save Too much time. Two years ago, an investor mentioned in an exchange with the author: Sometimes, to understand a track, reading one of your articles is as good as reading 100 reports written by institutions.
Why are there high-level and middle-level self-driving companies among the active users?
Because one of the characteristics of the nine chapters is that it can write technical topics in such a way that "people who do the market don't find it profound, and people who do technology don't think it's superficial", and they can write business models and market situations. The theme is written "so that people who do technology don't find it profound, and people who do market don't think it's superficial."
Therefore, the high-level and middle-level start-up companies, whether they are in the market or technology, can read the articles in the nine chapters to make up for some of the shortcomings in knowledge or information.
The CEOs and CTOs of a number of self-driving companies that rank very high on their respective tracks told the author that they often send nine chapters of articles to employees as "textbooks". There is even a president of a listed company who discussed a certain article of Jiuzhang at the company organization meeting.
Why are autopilot product managers from OEMs among the active users?
Autopilot product managers have very high requirements on people’s knowledge structure. A perfect autopilot product manager should be one of the people with the most complete knowledge structure in the company. However, there are not many such people. In fact, every product manager They all found that they "still have a lot of shortcomings to make up for." Frequent reading of Jiuzhang's articles and occasional participation in Jiuzhang's offline salons are an important way for them to "make up for shortcomings".
After summarizing the above content, the author realized that the most accurate positioning of the nine chapters should be: to help marketers understand technology, and to help technical people understand the market.
3.
Why don't Jiuzhang's active users include these two types of people?
Earlier, when listing the active users of Jiuzhang, the author mentioned the high-level and middle-level executives of the autonomous driving company, but did not mention the front-line engineers (in fact, many front-line engineers do not even know who "Jizhang is"). Why did this happen? What kind of difference?
The reason is actually very simple:
Jiuzhang's positioning is "to help marketers understand technology, and help technology people understand the market." The company's top and middle managers, whether they are in the market or technology, need to make up for their shortcomings by reading Jiuzhang's articles plate;
Among the first-line engineers, only a few have the need for horizontal expansion (to understand other technical modules and understand the market), most of them are still willing to be a screw, and have no demands for horizontal expansion. Of course, he doesn’t need to read the part he’s working on (the part he’s working on, of course, Chapter Nine is not as professional as them, and they definitely won’t read it).
Why doesn't Jiuzhang's active users include the middle and high level of OEMs?
This has to start with the style of the nine chapters.
As early as three or four years ago (when Jiuzhang had not yet been established), the author found very depressingly that most of the articles/reports written by the media, securities companies and consulting companies on the autonomous driving industry were "too macro" and "too hollow" , This style of content is often only suitable for two types of people: laymen; higher-ranking leaders-the author seems to have "bad intentions" when juxtaposing higher-ranking leaders and laymen.
For those target users (laymen, leaders), these too macroscopic and too empty contents appear to be "easy to understand"; but for the real elites in the industry, too much of these contents appear to be "too high a nonsense rate and effective information The proportion is too low."
The author took the lead in exploring a writing method with "relatively fine granularity", "very low nonsense rate, and high information density", which makes the article have a strong "sense of technology", but it is difficult for users with poor reading ability or impetuous In the case of a relatively high proportion, such articles are considered "too readable". Therefore, compared with other articles on the platform, the open rate of these articles with a strong "technical sense" is relatively low .
However, the author always adheres to a principle when writing articles: resolutely not cater to readers who like to read empty nonsense, but to screen out readers who "read as well as me" by continuously outputting content with a relatively high reading threshold.
In the years before the establishment of Jiuzhang, the author has always adhered to this standard; after the establishment of Jiuzhang, the author has promoted this standard to other colleagues, and repeatedly preached "non-essential, no nonsense" as the core value of the team .
The result of this series of efforts is that none of the original articles in the nine chapters are not fine-grained, and none of them contain a high proportion of nonsense. Therefore, it is also "brain-burning" to read—— For most people who are not up to standard or impetuous, such articles make them "unable to read", while for the group of academic masters, such articles make them full of "reading pleasure".
Of course, Jiuzhang, as an industry think tank in the direction of to B, certainly does not just stop at the level of "reading pleasure" for the value it brings to Xueba users. During the auto show at the Shanghai Auto Show, after intensive exchanges with the founders and executives of several autonomous driving start-up companies, the author suddenly made a major discovery:
Overall, horizontally, the product sales or valuation of an autonomous driving company is positively correlated with the familiarity of the founders and executives of the nine chapters; vertically, within an autonomous driving company, a person's rank It is positively correlated with his familiarity with Jiuzhang (Jizhang's communication with start-up companies often starts with the founder).
Of course, the above mentioned "positive correlation", not "causal relationship". In essence, these companies and these people are not "awesome because they often read nine chapters of articles", but because they are not impetuous and have a strong ability to learn new things, so they are awesome; these awesome people Being able to "read" nine chapters of articles with fine granularity and high information density has a good perception, so the quality of decision-making is further improved.
In fact, when a friend asked "Who is your target user" before, the author often joked that it is a student master, to be precise, "a person with a relatively strong reading ability".
Having said that, the answer to the question "Why is XXX not an active user of Jiuzhang" is very clear - for these people, the reading threshold of Jiuzhang's articles is "too high".
4.
The "perception system" of the middle and high level of the OEM
Of course, from the perspective of the bosses of OEMs, "broken articles" with fine granularity and high information density cannot be "food". system” cannot function normally and cannot make decisions.
So, let's take a look at how the perception systems of these bigwigs work.
At the end of 2020, before Jiuzhang was established, the author attended a closed-door meeting organized by the Association of Automobile Manufacturers. It is something outdated, which in my opinion belongs to "common sense" and "doesn't need to be discussed at all".
After the meeting, the author talked about this discovery with a friend, and the friend said: the weather is going to change, they haven't realized what happened yet.
Last year, Jiuzhang held several self-driving technology salons. A friend asked if there were any OEMs among the participants. The author said: From the results of our statistics, only a few technical experts from OEMs will participate, and their positions are relatively high. People basically won't.
My friend asked the reason, and the author's explanation is: From my observations over the years, these people are mainly active on forums where the proportion of invalid information exceeds 95%. Small salons with a low tolerance for nonsense don't seem to fit their needs.
On the third day of the Shanghai Auto Show, at a closed-door salon that the author attended, the president of a well-known think tank’s automotive research institute even proposed that “autonomous driving should be developed incrementally or by leaps and bounds”. 2.5 years ago, it was considered an outdated question that "has reached a consensus and no need to discuss it".
What shocked the author was that the executives of the automatic driving department of several traditional OEMs present not only did not think this question wasted time, but even praised the host for "good question". The "correct nonsense" in the eyes of ordinary people is actually regarded as "sight-seeing" by a group of bigwigs. If the quality of perception is like this, how can the quality of decision-making be guaranteed?
Some time ago, a friend who was doing industry research in an investment institution within a certain system told me this sentence: In many cases, the higher the leadership level, the lower the demand for information granularity. In the context of this article, this sentence is completely established. In fact, the bosses of OEMs are often cut off by the so-called "top international consulting companies". One of the key reasons is that the granularity of the information they have is too rough.
The author once thought that the middle management of OEMs were not active users of Jiuzhang, because the articles of Jiuzhang were too long, and these bigwigs “didn’t have time to read”; Look, aren't they busy?
It is said that a professional manager of an OEM with only 500-600 people in charge is busier than the founder of a technology company with 1000-2000 people in charge. I would not believe it. In essence, this is a question of the sense of mission and initiative of professional managers.
The boss of a technology company with a strong sense of technology thinks that he needs to be further popularized (horizontal expansion); while the professional manager of a traditional car company with insufficient or no sense of technology thinks that he "does not need to be popularized"?
The people who need to learn autonomous driving most happen to lack the motivation to learn the most, and these people are still in key positions in their respective companies. The author really sympathizes with the technical talents who work under these people.
5.
"The level is too high", don't you need to understand such a detailed thing?
Some people may say: "The bosses of the main engine factory are relatively high-level, and they only need to understand the macroscopic things, and don't need to pay attention to the fine-grained things." This sentence really makes those nonsense who only understand the correct The bosses felt great comfort. But is it really so?
At the end of 2019 or the beginning of 2020, the author once heard a friend relay the evaluation of the head of Baidu’s autonomous driving business at the time on the “level of understanding of autonomous driving” by the bosses of major car companies. Among them, Wang Chuanfu said these four words when evaluating BYD : I understand everything. The so-called "knowing everything" naturally means not only understanding the macro, but also understanding many micro and fine-grained things.
Did Wang Chuanfu's "know everything" create any value for the company? Of course it was created.
Many people in the industry are curious about a question: BYD is so good, why is it so late to launch autonomous driving? Recently, the author got the answer from a friend who has done in-depth research on BYD: Unlike most OEMs who believe that intelligence and electric integration are "parallel relationships" and can be promoted simultaneously, they go to the electric vehicle before the electrification is ready. In terms of intelligence, BYD believes that "the first half is electrification, and the second half is intelligence", and there is a sequential relationship between the two.
Indeed, before the EE architecture of the vehicle is clarified, many of the investment in the direction of intelligence are actually invalid investments. It now appears that BYD was right.
Therefore, those aforementioned people who think that they are "high enough level, and only need to understand macroscopic nonsense, not fine-grained things", should seriously think about a question: Is your "level" higher than Mr. Wang Chuanfu ? ?
As I wrote this, out of curiosity, the author entered the names of the founders of major OEMs in the backstage user data column of the Jiuzhang Official Account and searched, and found that these relatively high-level people have basically never paid attention to Jiuzhang, but there is one person The exception is Li Xiang—in terms of time, Li Xiang should have paid attention to Chapter Nine because of a fine-grained technical article "Killing IPS, a key leap for Tesla's pure visual route").
Judging from the content of Li Xiang's usual speeches, the granularity of information he can digest is indeed much finer than that of top leaders of other car companies. When people in the industry mention Li Xiang, they often say that "the learning ability is super strong", and some people even say that Li Xiang's speech is "a PUA fellow".
In addition, the friends of Xiaopeng and NIO's autonomous driving departments that the author has contacted also believe that the ideal autonomous driving strategy is better than their own.
That's where it gets interesting. Among the traditional car companies, BYD is generally recognized as the best, but among the new forces, it is said that Ideal is the company with the strongest overall competitiveness. There should be no controversy by now. Why are the two most competitive companies exactly the two with the "finest granularity" of the leader's perception system? Is this full of mysteries?
Therefore, those bosses (including a large number of professional managers) who think "I am high-level, so I don't need to pay attention to such details" should really wake up. After all, people with a "higher level" than you have a finer granularity than you.
If you are still not interested in understanding the microcosm and don't pay attention to the fine-grained things, then you can only continue to be taxed on IQ on macro issues.
We often find that some media or consulting companies provide overly macro views, which are regarded as "correct nonsense" by real insiders, but are regarded as the bible by some OEM leaders. The reason is that you usually only pay attention to things with very coarse granularity, so you lack the ability to process fine-grained data, so you have to highly rely on other people's algorithms (viewpoints).
The finer the granularity and the closer it is to the original data, the less information is lost during transmission.
Except for this auto show inventory series, Jiuzhang seldom directly outputs opinions in his usual articles, because we believe that think tanks like to output opinions and give pointers too much, which is tantamount to "the default user is a fool", and we believe that Jiuzhang users With our own judgment, we only need to sort out the facts in a structured way, and users can draw their own opinions.
Therefore, we mainly provide "feature-level data" that has been purified from the original data. Those decision makers with strong algorithm capabilities use these data to perform "feature-level fusion" with their own data, that is, "medium fusion" - similar to What BEV+Transformer is doing.
In contrast, decision makers who only focus on the macro and only understand coarse-grained articles or reports are equivalent to a gap in their own perception algorithm capabilities. Target-level fusion", that is, "post-fusion".
Do you understand? Wang Chuanfu and Li Xiang, who have a strong ability to digest fine-grained information, have truly mastered the "medium fusion" algorithm similar to BEV+Transformer, and most of the so-called "big guys" only have The ability of "post-integration" has been outdated, so how can you not be beaten by others "you can't find north and south"?
Of course, in many traditional OEMs, the top leaders have been engaged in machinery for a lifetime, and they are also old. It is unlikely that they will have "software thinking". Reality. But they should at least require the middle management of the company responsible for decision-making on autonomous driving matters to have these capabilities.
If you think that the above analysis logic in this article is not nonsense, then you will accept this point of view: a person's promotion opportunities in the autopilot business system of the OEM should be linked to his "reading comprehension" of the nine chapters—— —Those who haven’t read the articles in Chapter Nine, or those who “have read, but can’t understand”, should not be placed in key positions at all; those who have already been placed should also be “rolled down”.
Of course, there are still measures to make up for it, it depends on whether you are the "careful person".
Reference material: An atypical story: a car company develops its own full-stack, and I graduated from the full-stack
https://zhuanlan.zhihu.com/p/635267498
END
The fourth video of Nine Chapters Observation is officially launched
Welcome everyone to pay attention , forward , like and comment! !
Communication group| Enter "Sensor Group/Skateboard Chassis Group/Car Basic Software Group/Domain Controller Group", please scan the QR code above, add Jiuzhang Assistant , be sure to note the name of the communication group + real name + company + position (no remarks Unable to pass friend verification)
write at the end
About Contribution
If you are interested in contributing to "Nine Chapters Smart Driving" ("knowledge accumulation and sorting" type articles), please scan the QR code on the right and add staff WeChat.
Note: Be sure to note your real name, company, and current position when adding WeChat, thank you!
Quality requirements for "knowledge accumulation" manuscripts:
A: The information density is higher than most reports of most brokerages, and not lower than the average level of "Nine Chapters Smart Driving";
B: Information needs to be highly scarce, and more than 80% of the information needs to be invisible on other media. If it is based on public information, it needs to have a particularly powerful and exclusive point of view. Thank you for your understanding and support.
Recommended reading:
◆ Nine chapters - a collection of articles in 2022
◆ "Your budget is too much, so we can't cooperate" - Jiuzhang Zhijia 2nd Anniversary Review (Part 2)
◆ What is the comprehensive SOA-based electrical and electronic architecture?
◆ Application of deep learning algorithm in automatic driving regulation and control