Dialogue with AI Daniel Zhou Ming: How many hurdles have to be overcome from a scientist to an entrepreneur?

Mingmin comes from Aofei Temple

Qubit | Public account QbitAI

Age and seniority are the assets of world-class scientists.

But when you start a business, you need to face the skinny reality.

Can I work overtime? Can you work hard all night? Can you still fight at 50? Are the achievements and capabilities of small teams competitive with large companies on the same track?

This is the real experience of AI tycoon Zhou Ming , and it will also be the common torture faced by all top scientists after turning around and starting businesses.

Scientific research achievements and skills are inseparable from the accumulation of time, but the venture capital market may be programmed to favor younger people.

From this perspective, in the new cycle of technological innovation, there are still many outdated customs that need to be broken.

Knowing the year of destiny, why did Zhou Ming set out?

Zhou Ming, a well-known name in the fields of machine translation, NLP, and AI.

In 1991, he received his Ph.D. from Harbin Institute of Technology, an important NLP town, and then joined Tsinghua as a teacher. Until his departure in 2020, Zhou Ming had been the associate dean of MSRA for many years.

Zhou Ming is also one of the most accomplished Chinese in the field of NLP today . He is not only the developer of the first Chinese-English translation system in China and the most famous developer of Chinese-Japanese translation products in Japan, but also the chairman of ACL (International Association for Computational Languages), the most advanced conference in the field of AI language in 2019.

Zhou Ming’s scientific research and technological achievements have blossomed in both industry and academia. Among the NLP achievements he participated in and led the team, more than 100 papers were included in ACL alone, and related technologies have also been applied to world-class products such as Windows, Office, Azure, and Microsoft Xiaoice.

From the academic perspective, Zhou Ming’s influence in the field of NLP also ranks among the top in the world. He is one of the scientists with the most published NLP papers in the past decade. According to the latest data from Google Scholar, its articles have been cited more than 34,000 times, and its H-Index is 98.

So when he devoted himself to entrepreneurship, the industry sighed:

Even world-class AI scientists have bowed to the game.

Just starting the year of starting a business, Zhou Ming has reached the year of “knowing the destiny” as the Chinese often say.

Why go out?

Zhou Ming replied: The time is right and the time is right.

In terms of timing , around 2020, under the influence of the technical paradigm of “pre-training + fine-tuning”, NLP has developed to a period where it can be widely implemented.

AI large models represented by BERT and GPT-3 can handle a variety of tasks, and their performance is beyond the previous level. This means that NLP technology has found a route to explore in dealing with fragmentation problems, and production efficiency will also be greatly improved.

Geographically , domestic enterprises are going through a wave of digital transformation, which contains many opportunities for AI technology to be implemented.

Five years ago, perception technologies such as computer vision have helped some enterprises to identify intelligent “skills”, which has spawned more expectations and demands for AI technology in the industry.

And NLP-driven cognitive intelligence, there is a rigid demand and a market, but it is obviously in short supply.

In addition, in the general environment, the encouragement of industry-university-research conversion and hard technology innovation is unprecedented, and conditions in all aspects are ripe.

Human and human factors, the core lies in oneself.

Zhou Ming has worked in Microsoft Research Asia for 21 years, and both individuals and teams have met the expectations of “mission achievement”.

When he arrived at MSRA in early 1999, Zhou Ming set his wish to have a “double first-class” goal:

First, work with colleagues to make the research group enter the ranks of the world’s first-class research groups in the field of NLP;

Second, together with all colleagues in the field of natural language processing, we will strive to promote the research of natural language processing in China and even Asia to enter the world-class level.

And as a result, it was indeed achieved.

At this point, the unquenched entrepreneurial fire in my heart burned more and more intensely.

Zhou Ming knew that once he started a business, he would bet on all his achievements, honors and achievements in the past, and this determination must be made. And compared to the “best entrepreneurial age” in the popular sense, he is no longer young.

But he also quickly saw advantages on the other side of the coin:

First, AI is still the cutting-edge frontier of hard technology innovation, and NLP is known as the “jewel in the crown”. Without time accumulation, there is no technical insight and accumulation. Time is a friend, time is a barrier, and his experience and accumulation have advantages.

Second, his resume and age have long passed the stage of starting a business for money and food and clothing. Many people’s pressure and troubles in middle age have already come to an end, and they can concentrate more and only fight for their dreams.

Moreover, Zhou Ming also has deep technical confidence. He knows how to sail through the mountains in a light boat in the hustle and bustle of racing with big parameters and large models.

So Zhou Ming set off and the company was named: Lanzhou Technology .

The technical route is determined: lightweight large model .

Zhou Ming explained that despite the great popularity of large models in the past two years, the training and deployment costs brought by tens of billions and hundreds of billions of parameters will still discourage enterprises.

So blindly seeking greatness is not advisable. How to balance excellent performance and low cost is one of the feasible directions for the development of NLP.

Based on this kind of thinking, “Mencius” under Lanzhou Technology was soon born, and once on the stage, it was a thousand pounds.

On the CLUE list, the authoritative evaluation benchmark for Chinese language understanding, “Mencius” has completed the previous record of 10 billion and 100 billion parameter models with 1 billion parameters, and has broken Tencent, Sogou, Huawei, Alibaba Dharma Institute, etc. The pattern of giant players taking turns to dominate the list.

And this kind of ingenuity is not only reflected in the technical route, but also in the density of manpower.

To complete such a global record, the Lanzhou team is only a few people – it is density and efficiency, not scale.

In fact, in January 2021, when Lanzhou started to incubate in the innovation workshop, there were only 5 people in the team. In addition to Zhou Ming himself, there were 2 employees and 2 interns.

The blockbuster of the Mencius model also makes industrial needs, solutions and productization more urgent.

Soon, with the Mencius model as the core technology, Lanzhou Technology launched the “Mencius Cognitive Service Engine”, which includes functions such as search, generation, translation, and dialogue, and can also provide SaaS products for vertical industries.

At present, the “Rendu” machine translation engine cooperating with Transn is based on the Mencius model, covering more than 20 vertical fields and supporting thousands of translators on Transn’s platform.

The smart copywriting product contentnote, which Lanzhou cooperates with Shushuo and its subsidiary Ronghui, is also based on the Mencius model. The product was officially launched in December last year.

In addition, they have also reached cooperation with Huawei Cloud, Flush, China Asset Management, etc.

In September 2021, Lanzhou won the top prize in the Beijing HICOOL International Entrepreneurship Competition.

Lanzhou Company was officially established on June 10, 2021. Now, more than a year after its establishment, Lanzhou has received orders of tens of millions. Scientist Zhou Ming’s business thinking has become more and more mature, and he has begun to get more People sure.

It is also very fast. In February 2022, the second round of financing since the establishment of Lanzhou Technology has been in place, with nearly 100 million yuan of funds, Pre-A round, jointly led by Lenovo Venture Capital and Eight Roads Capital and followed by Innovation Works.

Zhou Ming explained that the meaning of financing is to gather more talented people, develop more fields, and realize the goal of the company’s founding:

Explore the application of natural language understanding technology.

How does Zhou Ming view “the grievances of scientists’ entrepreneurship”?

Aggrieved?

As a world-class scientist with a reputation, he has been challenged by age, physical and technical barriers in the venture capital market.

Zhou Ming smiled and said, ” No grievances “, every industry has its inherent cognition.

And this is not an honor or disgrace for one person, this is a new challenge for the whole country.

Zhou Ming believes that in the past two decades, China’s technology Internet has indeed achieved remarkable success in entrepreneurship, which is the result of the combination of the right time and place and the combined effect, but it has also brought about the wrong perception that innovation can be “quick success”.

Business model innovation can indeed be very fast, and the exploration of technology models has matured abroad. By using a broader market, population, and operating efficiency and local advantages in China, we can gain advantages and achieve success faster.

But the pros and cons are also written down at the beginning. In the past few years, the endless stream of stuck necks and supply cuts have all suffered from the lack of original technological innovation.

Zhou Ming said that it is enough to learn such a lesson once . Today’s respect for hard technology innovation also hides the society’s reflection on the rapid success of business model innovation.

Innovative startups in the new era should be technologically globally competitive from day one.

After starting his business, he has met many investors and people with various backgrounds. Zhou Ming is used to being asked everything about “age”. He can understand the other party’s starting point and concerns, but he still wants to say that we have to take the road of hard technology innovation. , Starting from technological model innovation, it is necessary to say goodbye to inherent cognition.

Zhou Ming believes that it is not advisable to continue to use the 2C thinking to understand hard technology innovation.

He explained how hard technology entrepreneurship is different from three aspects:

First, entrepreneurs need to have experience.

When hard technology becomes the anchor point of the company’s development, the founder’s understanding of the technology and industry is related to the company’s development and destiny.

For example, Zhang Zhongmou founded TSMC at the age of 56. His technical accumulation and industry understanding in the semiconductor industry over the past few decades is undoubtedly one of the important factors for TSMC to reach its peak.

Second, entrepreneurship is no longer purely based on commercial landing and opportunities of the times. It is necessary to master leading technologies to survive.

Previously, Zhou Ming mentioned that although China currently has advantages in AI landing, this advantage is not eternal.

Hard technology innovation still depends on the grasp of basic capabilities, especially key algorithms. For example, in cognitive intelligence, interpretability, knowledge modeling, etc. have not yet been solved.

Thus, innovators need to follow a “two-step process.”

The first step is to quickly implement the existing technology and serve the enterprise. After getting feedback from users, solve the pain points. Let the company survive in the market first and make money.

The second step is to return to basic technology and core algorithm development, so as to lead the development of artificial intelligence technology, so as to form a positive cycle and gradually promote the company’s development.

This naturally leads to Zhou Ming’s third point of view, that is, technological innovation is no longer a model of rapid expansion, but a step-by-step development , which requires not only the persistence of technicians, but also understanding and patience from the market. wait.

In such an innovation cycle, scientists with top innovation capabilities undoubtedly have unique entrepreneurial advantages.

In recent years, there has been a surge of scientists running away to start a business.

For example, in the main creative team of Transformer, 6 out of 8 people choose to start a business or join a start-up company; domestic AI giants such as Zhou Bowen, Jia Jiaya, etc., have established their own start-up companies.

However, Zhou Ming said frankly that scientists’ entrepreneurship also has shortcomings.

A problem he repeatedly mentioned is that scientists need to learn more from the perspective of entrepreneurs when dealing with things “related to people”.

He feels that doing scientific research is a matter of “fighting against nature”. The only thing to consider is how to discover new problems and come up with more powerful algorithms.

However, as an enterprise, we must consider how the technology can be truly used, and we need to take into account many issues such as application landing and cost.

On the issue of this combination of lengths and shorts, scientist Zhou Ming also felt that the solution was clear, and he would find more complementary big men to enter the company. He has already made it clear in his heart that he will not stick to the form and name. “If there is a suitable person in the future, the CEO of Lanzhou can let him do it.”

People gather because of things, things happen because of people.

The trajectory of life changed twice by Kai-Fu Lee

Zhou Ming said that in his half-hundred years of life, teacher Kaifu changed his trajectory twice.

For the first time, in 1999 , he was an associate professor of Tsinghua University, with the honor and teaching position of the highest university in China. Under the inspiration of teacher Kaifu, he gave up the establishment and joined the Microsoft Research Asia, which was just founded at that time and had an uncertain future.

The second time in 2020 , I met Kaifu at an event. Kaifu asked Zhou Ming: What are you busy with recently? Zhou Ming replied: I am thinking of starting a business. Kaifu was surprised and delighted: I understand you, and I will support you unconditionally.

However, Kai-Fu Lee, who is also a scientist, was also worried that Zhou Ming, who had been in MSRA for a long time, did not know enough about the operation mode of marketization, so he provided Zhou Ming with a compromise solution:

First, as the chief scientist of the innovation workshop, he is exposed to entrepreneurship and the market, and has a deep understanding of the elements and steps of entrepreneurship, and then starts to incubate the core team and start the company.

Of course, those who know Zhou Ming, believe in Zhou Ming, and support Zhou Ming are not only Li Kaifu.

After the news of Zhou Ming’s entrepreneurship came out, a call from Lenovo Ventures quickly called.

“Hey, Mr. Zhou Ming, have you really started a business? We want to invest.”

Zhou Ming said that Lenovo Ventures understands technology, and investors who understand technology do not need too much explanation.

Still soon, Lenovo Ventures grabbed Lanzhou’s lead.