The first $1 billion financing of a Chinese large-scale model startup has appeared.
On February 19, it was reported that the dark side of the moon has completed a new round of financing of more than 1 billion US dollars, and investors include Sequoia China, Xiaohongshu, Meituan, Alibaba, etc., and the old shareholders of the previous round continue to follow. This is the largest single round of financing in China's large-scale model track since the rise of large-scale model entrepreneurship. The valuation of the dark side has reached about $2.5 billion in the month following this round of financing.
The Dark Side of the Moon was founded in Beijing in April 2023, and its founder, Yang Zhilin, is a top student who graduated from Tsinghua University. In less than two months after its establishment, the dark side of the moon has completed an angel round of financing of nearly 2 billion yuan, with well-known VC investments such as Sequoia China, Today Capital, Zhen**, and Lisi Capital.
In less than a year since its establishment, it has raised nearly 9 billion yuan and has a valuation of nearly 18 billion yuan. In June 2023, The Information selected the five companies most likely to become China's OpenAI, and the Dark Side of the Moon was among them.
Three Tsinghua students started their own businesses
Different from the "second entrepreneurship" of Internet bigwigs who are light years away, Baichuan Intelligence, and 010000 things, the dark side of the moon is a rare "amateur" entrepreneurship in China's AI large model track. As soon as the dark side of the moon is established, it can be supported by heavy capital, and the resume of the founder is bound to be extraordinary.
Yang Zhilin is a post-90s generation who graduated from Jinshan Middle School, a century-old famous school in Shantou, Guangdong. In middle school, Yang Zhilin was selected to enter the training class of the Informatics Olympiad, and won the first prize in the Guangdong Division of the National Youth Informatics Olympiad League, and was qualified to be sent to Tsinghua University. Despite being escorted, Yang Zhilin still scored 667 in the college entrance examination that year, far exceeding Tsinghua University's admission line in Guangdong.
After entering Tsinghua University, Yang Zhilin continued the route of academic tyrants. Yang Zhilin was initially admitted to the Department of Thermal Engineering at Tsinghua University, but in his sophomore year, Yang Zhilin changed majors to enter the Department of Computer Science.
After graduating from Tsinghua University in 2015 with the highest grade in his grade, Yang went to Carnegie Mellon University's Institute for Language Technology and received his Ph.D. in 2019.
During his Ph.D., Yang Zhilin became a leading figure in the field of AI, and he has co-published with many Turing Award winners, and published more than 20 papers in computer summits such as ICLR, NEURIPS, ICML, ACL, and EMNLP, and his research results have accumulated more than 17,000 Google Shcolar citations. In addition to his academic achievements, Yang also works at Meta's Artificial Intelligence Research Institute and Google Brain Research Institute.
In 2016, Yang Zhilin, a Ph.D. student, started his first business and co-founded Circular Intelligence as a co-creator, with the direction of using AI technologies such as NLP, voice, multimodality, and large models to create "sales technology" solutions. Since its establishment, Circular Intelligence has completed 6 rounds of financing, with investors including Red Shirt China, Boyu Capital, GSR Ventures, Zhen**, etc.
In 2021, Circular Intelligence cooperated with HUAWEI CLOUD to develop the 100-billion-level NLP model Pangu model.
Although Yang Zhilin is only in his early 30s, his qualifications to make AI large models in China are almost unmatched, and he has participated in the research and development of large models such as Google abroad and Pangu NLP and Wudao in China (Professor Tang Jie of Tsinghua University, the leader of the Wudao project, is Yang Zhilin's teacher). Yes, it directly crowned Yang Zhilin with the title of the first person in the post-90s generation of China's large model.
In addition to Yang Zhilin, the other two co-creators of the dark side of the moon are also worth paying attention to.
Zhou Xinyu, the second co-founder and head of algorithms on the dark side of the moon, is Yang's undergraduate classmate and good friend at Tsinghua University, and the two also formed a rock band together during college.
Zhou Xinyu joined Megvii after graduating from university and studied the mass production of algorithms. The third co-founder, Wu Yuxin, also graduated from Tsinghua University and Carnegie Mellon University, and worked at Meta's Artificial Intelligence Research Institute after graduating from university.
In 2023, large models will suddenly become popular, and VCs are also trying to find China's OpenAI, but domestic talents with relevant experience in large models are very scarce, and there are only a handful of young entrepreneurs who have really engaged in large model research and development and are willing to start a business. The team of the dark side of the moon can be called the "top match" of China's large-scale model entrepreneurship, and it is no wonder that it has been stepped through the threshold by well-known VCs.
We want to be a "super app" in the AI era
In October 2023, the dark side of the moon released Moonshot, the first large-scale model since its founding, and Kimi Chat, a smart assistant product equipped with this model. As soon as it was released, Moonshot attracted attention for its "support for up to 200,000 kanji inputs". Keep in mind that GPT4 only supports up to 2With 50,000 words of input, Moonshot is eight times as much.
There are two key words in the research direction of the dark side of the moon, the first is long text, and the second is to c.
Regarding long text, the dark side of the moon believes that the application effect of a large model depends on two factors: the number of parameters determines how complex the large model can support "calculations", and the amount of text input that the large model can receive (i.e., long text technology) determines how much "memory" the large model has.
Support for long texts opens up new space for the use of large models. At the launch of Moonshot, Yang Zhilin demonstrated the input of the entire "Three-Body Problem" and asked the AI to give a summary, and also demonstrated the input of 50 documents at a time and let the large model give analysis.
Yang Zhilin analyzed, "When we look at the history of computer system development, an inevitable trend is that they all start from the very small memory computer services, to the large memory services. So I think there will definitely be the same trend for large models, from large models with little memory now to large memory in the future. ”
Another key word on the dark side of the moon is to C, and Yang Zhilin said that the highest priority task of Moonshot AI now is to find the direction of products, technologies and markets on the C side.
The positioning to C shows the ambition of the dark side of the moon to be China's open AI. Yang Zhilin believes that the large model may be divided into two different camps, to B and to C, and "to C is an opportunity to become a super app in the AI era", so the dark side of the moon firmly joined the To C camp.
Yang Zhilin said that the vision of the dark side of the moon is to "provide individuals with more convenient, stronger, and more personalized inclusive products through AI", which is very different from the idea of "empowering the B-side" of some other large model head players in China. The dark side of the moon, a team composed of the post-90s, shows a very different spirit from other large-scale model startups led by several senior bosses.
The positioning of To C makes the Dark Side of the Moon more willing to focus on the product, and thus maintains a relatively small and lean team. With the landing of the second round of funding, the team of the dark side of the moon has reached 80 people. Several other companies with similar valuations in China have teams of more than 200 people.
Post-90s entrepreneurs are emerging
Many people say that the last round of hard technology entrepreneurship is mainly a stage for middle-aged people, and a large number of new hard technology listed company founders are post-70s or even post-60s. But recently, post-90s entrepreneurs are quietly coming to the center of the stage and becoming China's new entrepreneurial force.
In addition to the Tsinghua trio on the dark side of the moon, there are many post-90s entrepreneurs who have been highly sought after by capital in the past year.
In 2023, the hottest Chinese post-90s entrepreneur may be Guo Wenjing of PIKA. Before Open AI released SORA, PIKA was the hottest Wensheng **AI, and its funding list was almost half of Silicon Valley's. Guo Wenjing, the founder and CEO of PIKA, is a post-95 genius girl who was admitted to Harvard University in advance for an undergraduate degree, and then dropped out of Stanford University to start a doctorate.
In December 2023, Zhiyuan Robotics, a humanoid robot startup, completed an A3 round of financing of more than 600 million yuan, and many institutions such as Lanchi Venture Capital, Zhongke Chuangxing, CDH Investment, Changfei**, C Capital, Hillhouse Venture Capital, Lijing Innovation, Sanhua Holding Group, Cornerstone Capital, Lingang New Area** and Ginkgo Valley Capital entered the game. The founder of Zhiyuan Robot is Huawei's "genius boy" Zhi Huijun, who started his business in February 2023 and has less than 5 rounds of financing in less than a year.
In August 2023, AI pharmaceutical startup Shenshi Technology announced the completion of a new round of financing of more than 700 million yuan, with investors including Zhongyuan Capital, Heyu Capital, Zhengxin Valley Capital, Evergreen Scitech Delta and a number of industrial capitals. The founders of Shenshi Technology, Sun Weijie and Zhang Linfeng, are a pair of post-90s graduates from Yuanpei College of Peking University.
The most recent example is the private aerospace company Oriental Space, which announced on January 24 that it had completed a Series B financing of nearly 600 million yuan, and its valuation also came to about 6 billion yuan, becoming the latest quasi-unicorn of China's private aerospace industry. On January 11, the "Gravity-1" rocket independently developed by Dongfang Space successfully completed its first flight, becoming the world's largest active solid rocket and the largest commercial rocket in orbit by China's private aerospace company so far. Yao Song, the co-founder and co-CEO of Oriental Space, is also a post-90s generation who was sent to Tsinghua University.
Different from the wave of post-90s entrepreneurs in the mobile Internet era, these post-90s are almost all famous school students, and the direction of entrepreneurship is at the forefront of current technological innovation. The back waves of the Yangtze River push the front waves, and they are the "light of youth" in this era.