Tsinghua's Link Genius Boy: When Top Workers Start Their Own Business
Original video by Bilibili up master “Bao Bao Ba 2022”
Backup of the video on this site (25:58, 121 MB)
The following is the text transcript of AI voice recognition:
Li Bojie 00:00
My wife always says that I spend all my time on the computer and don’t have much time to accompany her…
Up master 00:10
Okay. First, introduce yourself to the audience in front of the screen.
Li Bojie 00:14
Hello to the audience in front of the screen, my name is Li Bojie, I was a Ph.D. jointly trained by Microsoft and the University of Science and Technology of China from 2013 to 2019, and I was a genius boy at Huawei from 2019 to 2023, working on data center networks and high-performance systems for data centers.
Up master 00:31
Let me add that Bojie actually has various titles, such as the title of Genius Boy, the Youth Class of the University of Science and Technology of China, and he has also published papers at several top international computer conferences such as SIGCOMM and NSDI. The level of these papers is so high that only nature and science are left if you want to select a conference or journal that is more powerful than these papers. It’s basically at the ceiling level, and Bojie has started his own business now, so this time I want to talk to Bojie about his views on the workplace and life.
Is Bojie’s start-up really attracting a lot of attention?
Li Bojie 01:11
No. Actually, we want to keep a low profile. Why do we want to keep a low profile? You see, people always say that start-ups, why do start-ups have advantages over big companies? It’s fast and privacy. The first is fast, I don’t need a long decision-making process to start doing this.
And the second is privacy, I have a project and a good idea, I can quickly implement it, without having to say that many people are watching, because some breakthrough ideas in big companies are actually difficult to implement, the reason is that many people will ask you why you want to do this, privacy means that no one will ask you whether what you are doing is reliable or not, after you have done it, is it a mule or a horse, take it out for a walk, right?
Up master 01:49
Can you briefly introduce in one sentence what you are doing now?
Li Bojie 01:53
Just seven words will do. Large model operating system.
Large model, we are now mainly some large language models such as ChatGPT, and the operating system is a middleware between the large model and the application, which solves a series of problems such as the high and uncontrollable cost of the large model, security and model evaluation. We may have two types of users here, the first type of users are equivalent to to B, that is, these application companies, they need to use large models, but we know that the current large models, whether it is OpenAI’s model, or other open source LLaMA, they actually have security problems.
For example, it can’t guarantee that the output is compliant, and it also has hallucination problems, that is, the output may be nonsense, we have a way to know whether it is nonsense or not, it may not be reliable when executing long process tasks, it may not listen to people’s regulations, it has cost problems, because now whether using LLaMA or OpenAI’s GPT, the cost is actually very high.
In addition, it costs us $10 for GPT-4 to read a paper, and Runway ML, a video generation tool, costs $95 to generate a 7.5-minute video. We have a lot of accumulation in terms of technology, such as improving the performance of a model by 2~4 times, which reduces costs.
Secondly, we can also use small models to achieve a similar effect to large models, by doing alignment and other methods, we can fully utilize the capabilities of the model itself, reduce its hallucination.
Up master 03:32
Have you ever defined success or failure in entrepreneurship, such as what level I have achieved, I think I have succeeded.
Li Bojie 03:41
From the perspective of investors, it’s very simple. If your project is acquired or IPOed, it’s a success. If your project goes bankrupt in the end, it’s a failure.
If it’s from my own perspective, if the project is successful and people use it, that’s success. He can’t solve these cost and controllability problems with the current model. If he uses my thing and it solves the problem, then there is a demand, right? Right?
Up master 04:03
Have you ever thought about what you would do after success, when your company starts to make a profit?
Li Bojie 04:08
I think just gradually scale it, actually you asked a very key question, just like we now have some initial customers, what we are thinking about now is how to turn this initial customer into a sustainable product, from customer to product is a big difference, still need to consider how many people will use this product, how big is its market.
Host 04:27
Have you ever considered giving up if you fail? Have you ever thought about this?
Li Bojie 04:31
I think at present, there will definitely not be a stop signal for failure in entrepreneurship. Even if this project cannot be done, you must continue to do the next project until you succeed.
Host 04:44
What do you think are the particularly important points in the whole process of this genius teenager, including your experiences in the workplace?
Li Bojie 04:51
I think it’s like this, all the genius teenagers who come to the company are ordinary employees. He is just a special offer when recruiting, just like Ali Star and others, and the media will not publicize those Ali Stars. As a genius teenager, the company will give you many special training opportunities, such as opportunities for high-level exchanges. I remember the boss told us at the time, what is most important to young people, it is not money or level, the key is what kind of project this person is doing. Your entire life plan is composed of projects. He doesn’t care whether this project is academic or industrial, whether it is management or technology, as long as this project can make an impact, can make a difference. When I first joined MindSpore, I was responsible for the generation of operators. Most employees may be responsible for a small part after they come in. When I first came in, I was responsible for a key module of polyhedron compilation, and I also led a team to do technical things.
Then later in 2020, I was responsible for the unified bus, which is a new high-performance network bus in our company. This project also brings together many very powerful experts in our company, such as our Liao Bo, the chief scientist in 2012, and Tan Bo, my mentor, who is also a very senior expert in our Zhongsoft. They personally led us to do some designs, which brought us to the forefront of computer networks and computer systems, and benchmarked the most advanced network and system technologies in the world.
Host 06:21
Do you feel like there is a feeling of being a direct line and a non-direct line, ordinary people who go in do not have the opportunity to show, do not have the platform to show themselves, will they develop slower?
Li Bojie 06:35
This is definitely not the case. We said at the beginning that the genius teenager is used when he is hired. After he comes in, he is no longer a genius teenager. He just gave you a chance to get into this project, but it does not mean that those who get into this project will give you a high start at the beginning. If other employees do a good job, they can also step up. We have many people in Huawei who have been promoted very quickly, and even more are those who have risen from ordinary employees or ordinary doctors.
Host 06:58
That is to say, even if you enter Huawei at a general level, or even if you are an undergraduate student entering Huawei, as long as you do some powerful things on your corresponding platform, you will always be discovered, right? It’s not that if I don’t get in, I don’t get the so-called higher, I step by step, I don’t have that feeling, right?
Li Bojie 07:18
Look at some of the highest executives in our Huawei, which can be found on the Internet, they are not the earliest entrepreneurs in the 1980s, many of them joined Huawei in the 1990s or even around 2000, which shows that his talent selection system is very fluid.
Host 07:33
That is to say, even if you are not doing well in this step now, you can actually get to a better place in the future, right?
Can you talk about why you chose to start a business?
Li Bojie 07:44
Actually, I think Huawei is still very good to me. The leaders of our department care about me very much. At the same time, the high-performance network I am doing here is very critical.
If I continue to do the network, I will definitely stay in Huawei, but why I want to go out and do big models, the first is that I started to pay attention to big models from last year, it is indeed a key development trend in the next 5~10 years, not only after ChatGPT came out, but also many related technologies have proved this point, and now there are indeed many people trying to invest in things end-to-end to make a thing successful, this is also a challenge for me, because I actually wanted to do a product end-to-end when I was reading for a doctor and an undergraduate, although I didn’t start doing it formally at that time, I still enjoyed this experience.
From my current technical experience, I am a pure technical expert, without any product or business experience, but I hope to exercise my ability in this area through such an opportunity.
Host 08:39
Who do you think is suitable for starting a business?
Li Bojie 08:41
I think it’s very simple to ask a question to judge, for example, now that big models are so popular, do you think big models are more likely to succeed in a big company, or are they more likely to succeed in a startup?
If your answer is a big company, well, you definitely shouldn’t be suitable for starting a business, because there are indeed many people, such as our internal people in Huawei, Alibaba, Baidu, they have all talked about it, they think that big companies have resources and cards, and at the same time they have application scenarios, business scenarios, and direct traffic entrances. Another group of people think just the opposite, they think that the next wave of applications that may be successful, not to say that they are based on existing applications to patch up and add a big model capability, but AI Native. It’s a bit like our mobile Internet, the first wave may be the original Internet applications, I Google to make a mobile Google, right?
The really popular mobile Internet applications, whether it’s WeChat, Meituan, or Didi Taxi, these cannot be separated from the mobile terminal capability of the mobile phone, without GPS there would be no these applications, we believe that there will definitely be many key AI Native applications in the field of AI in the future, but what can be commercialized, we are not very clear about this now. We believe that small companies can overtake in some scenarios and make this happen, which is an opportunity suitable for entrepreneurship.
Up主 09:57
At the time when Steve Jobs founded Apple, computers were rather an anomaly.
So from this perspective, including the start-ups of Tencent and Alibaba back then, it’s not about financing, people said it was like pyramid selling, but if you really agree with this kind of finding niche areas, finding these niche tracks to do some things and succeed easily.
Li Bojie 10:18
I think the most suitable for entrepreneurs is to find a very small niche, and form a monopoly in the niche.
Up主 10:25
Bojie, what do you think is the difference between starting a business and working for someone else?
Li Bojie 10:28
That is to say, doing a lot of things in Huawei is actually very similar to starting a business, doing a project from 0 to 1.
At this time, I also have to write a proposal at the beginning, and when I write the proposal, I am actually raising investment, that is, our department leaders have to review whether the investment is beneficial to Huawei, and whether Huawei can obtain commercial value from it.
I think the biggest difference between doing this in a big company and doing this in a start-up company is that the risks are different, that is, if I am in a big company, if my project does not get investment, I can go and do other projects.
That is to say, but in a small company, you have staked all your time and all your assets, right? To do such a thing, so he is different in terms of risk and return. If it is a small soldier and an entrepreneur in a big company, the difference may be quite big, but if I have a certain position in the company and need to find resources for the team to do this, it is a leader of a thing, these two roles may not be the same.
Up主 11:24
I want to know, for example, what time do you usually get up in the morning and what time do you go to bed at night?
Li Bojie 11:29
Actually, I usually get up at around 8 in the morning and go to bed at 12 at night, which is quite regular. When I was in the company, I was actually one of the people in our department who worked relatively long hours, not because I was rolled, but because I liked computers. And my wife always says that I spend all my time on the computer and don’t have time to accompany her, so sometimes even at home, I am thinking about these things and like to spend some time on them.
Up主 11:56
If you were to do your PhD again, what advice would you give to yourself back then?
Li Bojie 12:01
First of all, I think my PhD was relatively smooth, because I happened to meet two supervisors. The first supervisor was very good to me, treated me like a family member, and fought for many advanced resources, including the most advanced network equipment in the world at that time, and some research clusters. So this is the first piece of advice, that is, you must go to a place where you can really get the most advanced applications, or the most advanced hardware, to do this PhD, of course, if you have already chosen, there is no way, but if you have not chosen yet, make this choice. In our field of computer science, many times it is closely related to the capabilities of hardware and its applications, that is, whether you can find a suitable application scenario, the first project he gave me was a hand-in-hand mode, I think this is also very good.
Finally, find a supervisor, that is, he is personally leading such a project hand in hand, because now there are many hands-off types, that is, free-range, that is, my second supervisor may be slightly more free-range, but the advantage of the second supervisor is that, because the idea of my first supervisor was proposed by him, and including the writing of the paper, he helped me change a lot of things, and including the process of the experiment, he was leading me step by step, so that I knew how to do scientific research, as if there was a template.
And then after the first paper was published, strictly speaking, I was the one who did the work, in fact, he should be the first author, but he is to let me graduate, right. Then to the second paper, the hand-in-hand teaching form is also not conducive to cultivating students’ ability to do independent research, but I just changed to the second supervisor, this is also very lucky, his interests are very broad, and then let me say you want to think about your own research direction, of course, he also gave me a lot of guidance in this research direction. His experience is very rich, at the beginning some not very reliable directions were rejected by him, and then let me focus on this problem, just happened to be interested in it, and then have relevant background knowledge, so that I learned the ability to do independent research later.
Up主 13:56
Who do you recommend to do a PhD?
Li Bojie 13:58
I think the most recommended type of person is actually someone who has a very strong interest in science, interest needs to add a limit, he is not the kind of interest in pseudoscience, not to say that I read a bunch of popular science articles, feel it is very interesting, such people often when they really encounter those trivial things in scientific research, then he will feel different from the actual imagination. For example, your paper is always rejected, right? Then it collapsed, so it’s best to have actual scientific research experience.
It’s best if I want to do a PhD, it’s best during undergraduate, he joins some laboratories to do some auxiliary work in scientific research, so that he has a preliminary understanding of what scientific research is doing, what kind of process it is, and then say whether I really want to do scientific research, if I still think it is of great interest to him, I hope to explore the frontier of science, and I am willing to say that I am willing to advance a small step in the frontier of science, then I think advancing this small step is very valuable, he is suitable for a PhD, in fact, many PhDs he is actually not suitable for a PhD some utilitarian ideas, I read this PhD is to earn more money, or I read this PhD is for a big achievement, right? But these are actually not very realistic.
Up主 15:02
Do you have any experience in paper submission or writing papers?
Li Bojie 15:07
Actually, my ability to write papers is not good. The reason why my first paper was accepted at once is because my tutor helped me revise it, and the whole project was led by my tutor from beginning to end, right?
The second project was able to be accepted at once, also because the first one was successful. I did the follow-up work on that basis. At that time, FPGA was being deployed in Microsoft, which was a world-renowned event. Writing a paper, in fact, the reviewers were very eager to see this paper accepted. There was such a platform to leverage. But it wasn’t so easy for me to publish papers later. Like now, I have submitted articles four or five times and they haven’t been accepted. The two papers I published later were also accepted after four or five submissions. So it’s really not that easy. The important thing is, first, to find someone who is proficient in paper writing to help polish, that is, to improve the article.
Up 15:56
That is to say, it is suggested that everyone should show their articles to some big cows they can recognize and let them have a look.
Li Bojie 16:04
Especially before this article has not started to write, when things are still in the conception, to the academic conference, and then communicate with these big guys, because the big guys are basically your future reviewers, to see if your idea is reliable, if everyone thinks your things are nonsense, you don’t write it, right?
Up 16:22
Can you talk about the most important choice in your life?
Li Bojie 16:26
That is, in the first semester of my sophomore year, I switched from mathematics major to computer science major. This started from college. At the earliest time, I was admitted by competition, so Jiaotong University and University of Science and Technology of China gave me admission, including Fudan University. I won awards in both computer science and mathematics, so I thought whether I should study mathematics or computer science. Then I felt that my mathematics might be better. I wanted to calm down and read it. At that time, I didn’t understand advanced mathematics enough. I felt that it was similar to the competition mathematics in junior high school. I went to read Hua Luogeng class, our Hua Luogeng class in University of Science and Technology of China, because the mathematics science in University of Science and Technology of China is relatively good, right?
But after reading for a year, my math grades were not very good, and my interest in math was relatively poor. I felt that they were pushing these formulas every day, and I felt a little bored. At the same time, I was very interested in computers. After I went to college, I came into contact with a lot of website building, including operating system building, including network maintenance. I felt that I could really do something from beginning to end. I was very interested.
So in the first semester of sophomore year, I switched to computer science. I think if I didn’t switch at that time, I might not be able to become a mathematician. I might be more painful because I was doing something I didn’t want to do. So I think the most important thing in college is to find something you are interested in and dare to do it.
Up 17:37
OK, I think Bojie has a very good quality. Before you do something, you will try it first. Including what you just said about recommending what to read for a doctor, everyone says what’s not good, you persuade him to say you try, you try first, maybe this thing is very simple, they don’t even have the courage to try this step, because they go to the laboratory to try during their undergraduate period, including what you just said about why you choose to switch from mathematics to computer science, also because you go first, you feel that you are at that time node, you feel that you like mathematics very much, but after you study, you find that it seems to be just like that, so you turn around and jump to another one, I think on the contrary, I think it’s a very big factor that you can be so powerful and successful, because you dare to think and do, really dare to think and do.
What do you think the help of your original family is?
Li Bojie 18:30
Actually, I feel that my family earns less money as a wage earner, and at the same time, I feel that among the people with similar grades and the same level as me, my family condition is relatively poor. So why can I have such an opportunity to compete or something? The main reason is that I think the public education in Shijiazhuang city was well done at that time. That is to say, most of the schools in my era had to pay a lot of money to attend these competition classes and receive good education. But at that time, our Shijiazhuang No.2 Middle School was actually the best middle school in Hebei Province, and it gathered the best teachers in Shijiazhuang city, and all the competitions were free, including the math Olympiad competition we held in primary school, the free classes, the resources you got in public education were the best in Hebei Province, so this condition may not exist now.
Up 19:23
This is also a kind of luck, right, it’s not the kind of family can give you, right, at that time you didn’t say in the blog that there was an opportunity to take the normal college entrance examination, and then test Tsinghua, right? Then you didn’t choose, I saw in your article that you regretted it. What impact has this had on your subsequent life?
Li Bojie 19:42
The resources that Tsinghua can provide to me are definitely a lot, because for example, Tsinghua has a lot of alumni resources, whether in the academic world, our network circle of academia or in the field of entrepreneurship, in fact, Tsinghua has a lot of alumni, so it is definitely able to provide a good resource, but I also have its unique advantages in USTC, because although USTC has relatively few people engaged in computer science, but the school gives us more support, such as the school network center, then he gave us a lot of machines, let us build websites on it, provide some network services on it, many schools actually do not give students voluntarily.
In addition, our USTC group is more concentrated, we these technical people all know each other, and the people I cooperate with, then are all from the USTC circle.
Up 20:28
So there’s actually no regret, each has its own advantages and disadvantages. You think you’ve been through so long in your life, what period of time do you think was the low point in your life?
Li Bojie 20:40
Maybe it’s the first year or two of my undergraduate studies and the matter of changing majors. At the beginning, I really didn’t like studying mathematics. So I didn’t have much interest in these basic courses such as mathematics and physics. I was interested in the professional things of computer science, so my grades were not very good. I felt that this was a low point. But fortunately, after I joined Microsoft Research Asia, it was better. I did a lot of interesting projects when I was in USTC. Although my grades were not very good, the tutor of Microsoft Research Institute saw the things I had done before, because he was doing the same kind of system things, which was very similar to the things I had done with computer network and operating system before. So I performed very well here, and I quickly adapted to it. He gave me a good opportunity, and then I developed better later.
Up主 21:24
I feel like you’ve been listening quite smoothly. In our group, no one was leading. I was the one who sorted out the first wave of the first game. I barely made the line look a bit decent, but after a few days, it was messed up again. The most awkward thing is when you’re running a program smoothly, and then someone comes over and kicks your system off, ruining your entire testbed, and then you can’t do anything about it.
Li Bojie 21:50
Why did they ruin it? Weren’t you supposed to be allocated?
Up主 21:53
Our area is even more chaotically managed, no one is in charge.
Li Bojie 21:58
So whoever comes, if I don’t like the look of the machine, I’ll unplug it.
Up主 22:02
After encountering this kind of thing a few more times, everyone will ask in the group like ALOHA.
Li Bojie 22:10
Is anyone using it? If no one says they’re using it, I’ll take it, right?
Up主 22:15
Often that person sees it.
Li Bojie 22:17
The machine has already been taken. It’s a good thing if the machine is taken, but the environment has already been destroyed.
Up主 22:23
The environment is destroyed, and then the code is gone.
What do you usually choose to do when you’re under a lot of pressure?
Li Bojie 22:29
Actually, when I’m under a lot of pressure, I usually just read a book, or I go online to look at some materials and stuff. Not academic materials, I still like reading.
And then I also exercise, I run 10 kilometers every day, and I can run 300 kilometers in a month. So I’ve slimmed myself down, and at the same time, I used to fail in school, but later I was even able to run a half marathon. Actually, I think it’s because I was transitioning from an individual contributor at Huawei to a managerial role in an organization. Although I lead a technical team, I don’t need to directly evaluate them, I just need to give them some advice and work together. But as a real organizational manager, you need to consider more things, like planning the project, finding the resources needed for the project, and finally turning the project into a product. You need to think through the whole planning process, find application scenarios, and think about what technical competitiveness the project has, and come up with key competitive points.
Up主 23:33
One last question, do you have any words, any advice, for our audience?
Li Bojie 23:38
For myself, there’s the professional aspect and the non-professional aspect. First of all, like the current large models, it’s a very key technology for the next 5~10 years. So anything related to large models, whether it’s the basic infrastructure of the model, like hardware, its training framework and network, and communication storage and so on, and also the middleware between these models and applications, which is what we just talked about, the operating system, there’s a lot that can be done. We also feel that doing the most critical things in line with the trend might make it easier to achieve some results.
The second thing is the non-professional aspect. No matter what stage everyone is at, the most important thing is to find your interest and follow your heart. This also solves what you were saying at the beginning, that many people are just following the crowd, being a small soldier, and a lot of what small soldiers do is homogenized, right?
For example, a simple example is the large model, and now OpenAI has come out with a powerful large model called GPT-4. Now so many companies have come out and said they want to do something like GPT-4. In fact, it’s a me too thing, which might not be that interesting. What you really need to do is to figure out what makes you different from others, a bit like what Steve Jobs said in his famous speech, make a difference, right?
Up主 24:49
Okay, that’s it for today, say goodbye to everyone.
Li Bojie 24:52
Okay, goodbye, thank you.
Up主 24:58
Do you have any advice for engineering boys chasing girls?
Li Bojie 25:01
Actually, I can’t say I have any advice. You say do I have any experience, I don’t have any experience. I don’t think I have any experience in chasing girls. The most important thing is what?
That is, when you meet the right person, you should boldly pursue her, and always maintain a good relationship with her. And I myself actually also don’t do this very well. That is, many times the thinking mode of engineering boys is quite different from that of most girls. That is, boys’ thinking mode may be more rational. For example, the most typical example is that boys sometimes get angry, and then I will naturally get better after a night, and never overnight. But girls are different. If you don’t solve emotional problems in time, they will get worse and worse, right? This is a difference in thinking patterns. The key is to find someone who shares your values. I think this is very important. If you just think someone is pretty, but your values don’t match, it’s definitely uncomfortable to live like that.