Big model startup has added a new talent, Nai Xue’s former tea CTO also joined the team

Original source: Rongzhong Finance

Author: Zhang Congxiang

Image source: Generated by Unbounded AI‌

In the entrepreneurial boom of big models, another Internet veteran is welcomed.

Recently, some media broke the news that Naixue’s tea CTO He Gang has resigned and may participate in large-scale startups next. Subsequently, some media sought confirmation from Naixue’s tea official and confirmed that He Gang had resigned. Public information shows that He Gang joined Naixue’s Tea in June 2020 and is responsible for the company’s digital strategy and supervision of information technology management.

Artificial intelligence has always been a hot topic for entrepreneurship, and the iPhone moment triggered by ChatGPT has completely ignited people’s enthusiasm for large models and large model entrepreneurship. Data released by the Tianyancha Research Institute shows that as of the first half of 2023, there have been an average of 48 financing events per month. In terms of specific months, the number of financing events in May was the highest at 61, followed by 54 in June and March. From 50. The increase in the primary market has also led to the birth of various large models. According to the Ministry of Science and Technology’s “China Artificial Intelligence Large Model Map Research Report”, China has released 79 large models with a scale of more than 1 billion parameters.

On the other side of the wind, the battle between hundreds of models has made the track increasingly crowded. From a foreign perspective, Jasper, the first large-scale AI model unicorn, announced layoffs, and Mutiny, which completed a $50 million Series B round of financing last year, also laid off 30% of its employees; the domestic AIGC star company Movie Book Technology was exposed for poor management, and Wang Huiwen temporarily laid off employees due to physical reasons. Don’t be light years away. Or layoffs or struggling to survive, will the myth of large-scale entrepreneurship be broken?

**01 He Gang chases the trend, what to do with Nai Xue’s tea? **

On July 17, 2020, news broke that He Gang, the former Chief Technology Officer (CTO) of Luckin Coffee, will soon join Naixue’s Tea as CTO and will officially join the company in August. He Gang’s arrival is to help Naixue’s tea products prepare for further launch.

Time flies, and two years later, He Gang chose to leave and devote himself to large-scale model entrepreneurship.

Public information shows that He Gang was once a member of the junior class of the University of Science and Technology of China. In 2011, before returning to China to join Shanda Cloud, He Gang worked at Microsoft and Amazon for 12 years, responsible for cloud computing related technologies. After joining Shanda, He Gang and Ji Xinhua, the former chief security officer of Shanda Online, co-administered the management as co-CEOs. He served as CEO of Shanda Cloud Computing and vice president of Shanda Group. In March 2012, He Gang resigned from Shanda Cloud. In April of that year, he joined JD.com as vice president of technology and chief scientist. In September 2019, He Gang joined Luckin Coffee. His main responsibility is to manage the information center. The technical directors and data directors of each business line must report to him. After the financial fraud scandal broke out at Luckin Coffee, He Gang became the first senior executive to leave Luckin.

While serving as Naixue’s tea CTO, He Gang promoted the digital construction and upgrade of Naixue’s tea front-end and back-end, including the upgrade of online channels such as front-end ordering mini-programs, back-end management of self-developed digital operation systems, and the use of digitalization to drive product innovation. . During the epidemic, Naixue’s Tea also relied on digital construction to survive, relying on small program ordering, access to third-party takeout platforms, and online micro-malls, live broadcasts, Tmall flagship stores and other online channels to accelerate online and offline integration. , to create digital new retail. These measures also allowed Naixue’s Tea to achieve a surge in the number of members in 2020. Even Peng Xin, the founder of Naixue’s Tea, said at the time that Naixue’s Tea had started to build its own membership system in the second half of 2019, with an average of 2 million new members added every month, and the repurchase rate was very high.

Nowadays, Naixue’s Tea has just started in terms of digital links. He Gang’s departure will more or less affect the advantages accumulated by Naixue’s Tea in supply chain and digitalization.

In Peng Xin’s plan, the future digital direction of Naixue’s Tea is mainly reflected in the supply chain, store operations, business management, etc., through a system plus automated and semi-automated equipment to replace the original manual repetitive work. The role of digitalization is to simplify Naixue’s overall business processes. For example, from the clerk making tea to the semi-automatic machine making tea, from letting people remember the formula to the active reminder of the system, from the store manager’s experience-based distribution to systematic decision-making, based on these digital systems, Naixue’s overall tea management and Execution becomes standardized and intelligent.

It can be seen in the recent financial report of Nayuki’s Tea that the high-quality and cost-effective product strategy has been implemented. In addition, the three major cost rates of raw materials, rent, and manpower catering operations have dropped by 8.6% year-on-year. Therefore, Nayuki’s Tea can Turn losses into profits. Among them, the labor cost rate at the store level is stable within 20%, and the actual rental cost rate remains within 15%. As costs continue to be optimized, store profitability is expected to continue to improve. As of June 30, Naixue’s Tea had opened 1,194 directly operated stores in 93 cities across the country, with a net increase of 126 stores and approximately 66.4 million members.

It is undeniable that while working at Naixue’s Tea, He Gang did help Naixue achieve remarkable results in cost reduction and efficiency increase from preparation for IPO to today through a series of digital changes. Based on his past experience in data models, He Gang left Naixue to join large model entrepreneurship, which will also arouse people’s attention and expectations in the field of artificial intelligence.

02 Funds and talents are the two pillars of large model entrepreneurship

When big models became the trend, Internet veterans such as Meituan Wang Huiwen, Sogou Wang Xiaochuan, and Kaifu Li began to join the game. Later, domestic Internet giants Baidu, Alibaba, SenseTime, Byte, etc. followed suit, including many young people. Entrepreneurs have joined, such as Huawei’s young geniuses Zhihui Jun and Li Bojie, who have also participated in large-scale entrepreneurial ventures.

Behind the rapid growth of large-scale models, we need sophisticated industry talents and the most indispensable funds. “We estimate that the training cost of a general large model is about 1 billion US dollars. This is only the computing power part, not counting the other two very expensive parts, one is data and the other is labor cost. Now the global large model Talents in this field are very scarce.” Dr. Du Feng, founding partner of Jiangmen Ventures and former head of Microsoft Ventures Greater China, once said publicly.

And indeed it is. As a research and development company of ChatGPT, OpenAI consumes 700,000 US dollars every day to operate ChatGPT, which requires continuous large-scale investment. This is a huge capital expenditure for OpenAI, which does not have high returns, so much so that OpenAI CEO Sam O In the second half of the year, Terman went to Abu Dhabi, the capital of the United Arab Emirates, and other places to seek financing, and the scale of financing was also huge, with an amount of no less than US$100 billion.

In fact, OpenAI has completed a round of tens of billions of US dollars in financing this year, and its valuation has reached nearly 20 billion US dollars. Venture capital companies participating in this round of financing include Tiger Global Management, Sequoia Capital, California Andreessen Horowitz, New York Thrive and K2 Global. But on August 10, Indian media Analytics India Megazine reported that if Sam Altman continues to burn money at the current rate, OpenAI may be forced to file for bankruptcy by the end of 2024.

In addition, according to US media reports, due to the development of ChatGPT, OpenAI’s losses in 2022 have approximately doubled to US$540 million, but its revenue is only US$3 million. OpenAI CEO Sam Altman’s goal of revenue of US$200 million this year and US$1 billion next year will also be difficult to achieve.

This is the case for leading large-scale startups, and it is even more difficult for domestic entrepreneurs entering the game. Guosheng Securities once estimated that the cost of training GPT-3 once is about US$1.4 million. For some larger large-scale language models, the training cost is even between US$2 million and US$12 million. For such a high-cost game, without continuous cash input, it will be difficult for large-scale startups to continue.

On the other hand, talent is also an insurmountable gap in the large-scale entrepreneurial track. From Wang Huiwen, the co-founder of Meituan, who raised his voice two months ago to leaving his job and seeking medical treatment due to depression in the later period, the story of building a Chinese version of OpenAI ended here. The reason why investors cleared their positions one after another is that they valued Wang Huiwen. This person values his ability to integrate resources. However, the capital market is realistic. Once the link of people’s absorption is lost, the company will also face the dilemma of shrinking valuations light years away. From its valuation of US$1 billion just a few years after entering the market to its final acquisition by Meituan for US$285 million, the myth of big-model entrepreneurial stories seems to have begun to be shattered.

Large models are a hard technology track with high requirements on computing power, algorithms, and data, and behind these technologies, there is a need for corresponding talent support. From the perspective of the recruitment market, the “AIGC Talent Supply and Demand Report for the First Quarter of 2023” released by Lagou Recruitment shows that in the first quarter of 2023, the demand for AIGC talent recruitment has increased for three consecutive months. In March this year, the demand for AIGC talent positions increased by 42% month-on-month. . On recruitment platforms, many companies even offer millions in annual salaries to compete for AI technical talents.

The poker table of large models is gradually becoming crowded. Whether they are the first to enter the game or entrepreneurs who want to participate, facing the two mountains of talent and funds, it has also become a key factor in widening the gap among large model entrepreneurs.

03 The track is hot but the capital is cold, and the myth of large models is broken

Although the industry shows endless enthusiasm, judging from the financing in the primary market, leading VCs are keeping their wallets tight. How can the myth of large-model entrepreneurship be sustained?

At this year’s World Artificial Intelligence Conference in Shanghai, more than 30 large models were on stage, and not many companies were able to obtain more financing amid the excitement. According to media statistics, there are incomplete statistics on the financing of AI large models in the first half of the year. There were 107 financing events in the field of AI large models in the first half of this year. Less than a hundred AI startups have received financing, and the financing amount is generally in the tens of millions. to hundreds of millions. The amount of financing for large models is also much higher than that of technology hot spots in previous years, such as autonomous driving, smart logistics, lidar and other artificial intelligence projects. From a larger perspective, global venture capital funding almost halved in the first six months of this year, falling 48% to $173.9 billion, and the number of deals also dropped 19%, according to research firm PitchBook. In 2021, global venture capital funds hit an annual record of US$745.1 billion.

This has created a strange situation: the track for large-scale model entrepreneurship is very hot, and people are rushing to it, but VCs who can invest in it are relatively calm. Even the leaders who were not stingy in the field of AI in the past VCs, including Sequoia, Hillhouse, IDG, Wuyuan Capital, etc., also made fewer transactions.

This is because leading VCs have grasped new trends earlier, and VCs with a keen sense of smell have already completed the layout in advance. When new entrepreneurs come in, they naturally need to observe and consider them to a certain extent. Chapter 42 founder Qu Kai revealed that almost all US dollar funds are currently looking at AI, and some RMB funds are also interested. “Many institutions are very positive, but not many will take action in the end. There will be about fifty or sixty.”

The entrepreneurship trend of large models has to a certain extent caused false prosperity in the industry. This is because large models themselves face factors such as large investment amounts, long return cycles, and low success rates. In addition, coupled with the dominance of large manufacturers such as BAT, To a certain extent, this determines the final fate of large model entrepreneurs. On the other hand, all parties in the industry have not reached a basic consensus on the entrepreneurial value of large models. For example, Zhu Xiaohu, managing partner of Jinshajiang Venture Capital, and Fu Sheng, chairman of Cheetah Mobile, have publicly quarreled with each other over the entrepreneurial value of large models.

The point of difference between the two is also whether the general large model has prospects or the vertical large models have prospects. In fact, the current main view in the industry is that the underlying large model itself is changing. In the future, it is likely that there will be a large number of open sources, or 1-2 leading manufacturers will achieve winner-take-all.

The current commercial implementation of global large-scale models is still in the early exploration stage. On the one hand, R&D institutions are familiar with the basic technologies of large models, but lack effective implementation scenarios and need to cooperate with downstream scenario companies to establish large model business models. On the other hand, most companies in downstream scenarios do not have enough computing power and development resources to customize or fine-tune large models.

Faced with such a powerful gold-eating beast, the imagination that large models bring to people seems to be far less clear than its specific future business model. Everyone knows that the commercialization of large models will be very strong, but how to achieve commercialization has become a major issue in entrepreneurship. problems between investors and investors.

Although OpenAI has taken the lead in adopting an open strategy, cooperating with many technology companies and research institutions to develop and promote technology development to achieve resource sharing and win-win results, and at the same time trying to implement applications in many fields such as medical care, finance, education, etc., how to advance In the process of technological development, maintaining one’s own influence and hematopoietic ability are the basis for sustainable development.

Regarding the later development and industrial application of large model technology, we can see from the fact that several large model giants are beginning to consider sporadic commercialization. The future development trend of large models is no different from the past history of Internet technology. Bottlenecks and industrial explosions continue on the path.

Perhaps, for large model companies that are trying to cross the river by feeling the stones, it is time to rethink the feasibility of their business models.