It’s about contracting a forest. Dark Side of the Moon founder Yang Zhilin: Artificial intelligence entrepreneurship is not about planting a tree
In March last year, MoonshotAI, a company focusing on the research and development of large-scale general artificial intelligence models, was established. In October, it launched Kimi, the world's first intelligent assistant product that supports the input of 200,000 Chinese characters, which caused a huge response on both the user end and the capital market. . At the 50th Shanghai Innovation and Entrepreneurship Youth Forum held today, Yang Zhilin, founder of Dark Side of the Moon, shared his entrepreneurial experience and insights.
Few people know that behind his concentration on hard technology, this talented student’s childhood dream was to be a rock singer and wandering poet. He said: "The spirit of innovation is the same as the spirit of rock and roll. It uses new things to bring new increments. When computers first began to develop, IBM was the largest company with the highest market value in the world. People described IBM as both the sun and the moon. It was the star again, and the sun, the moon, and the stars were all covered by it. At that time, there were only a few people at Microsoft, but they still made disruptive innovations. This may be a direction we want to work on.”
Yang Zhilin is moving forward along the entrepreneurial path he envisioned - Dark Side of the Moon, which was established last year, has now become one of the six unicorns in the field of general large models in China. On February 19 this year, Dark Side of the Moon completed its latest round of financing, amounting to more than US$1 billion, setting a new record for the largest single financing in the domestic AI field. After this round of financing, Dark Side of the Moon’s valuation has reached approximately US$2.5 billion.
Yang Zhilin was born in 1992. He graduated from the Computer Science Department of Tsinghua University with a bachelor's degree. He holds a PhD in computer science from Carnegie Mellon University. He studied under the former AI director of Apple. He has worked at Facebook AI Research and Google Brain. He is the most cited person in the field of natural language processing under the age of 35 in China. The highest researcher.
When talking about the reasons for returning to China to start a business, Yang Zhilin said that one of the main reasons was the complete environment. Whether it was policy support, venture capital support, or the accumulation of domestic related industries and talents, he saw good entrepreneurial opportunities. The other is the rapid development of AI itself. He believes that as the amount of computing increases exponentially, it is very likely that AI products within everyone’s reach will appear.
"The development of technology is very magical. Why can GPT-4 released at the end of 2022 become a milestone? Because it can do things that were impossible before." Yang Zhilin said that this is actually the result of the right time and place. The development of the Internet It has created a lot of value for more than ten years, but it is very likely that when we look back in thirty or forty years, we will find that the greatest value of the Internet is the accumulation of large amounts of data for AI. In addition, the emergence of AI also benefits from hardware advancements, including the proposal of the Transformer architecture and the development of semiconductors.
So what is the next new development direction of artificial intelligence? In Yang Zhilin's view, it is necessary to develop from answering simple questions to solving complex tasks with long links, and also from completing 1% of the workflow to 90% of the workflow. "For example, based on some very professional documents, all your subsequent learning processes are carried out in artificial intelligence, and you can quickly become an expert in a new field. This is also a very important application scenario."
What exactly is the ability to deal with more complex problems? Yang Zhilin explained that for example, when using artificial intelligence to analyze a research progress, it must not only know the latest progress and which research institutions are doing well, but it can even directly write code, analyze experimental results, and discuss the next steps with the experimenter. research direction. "That is to say, it can not only passively follow your instructions, but can also challenge you. You give it an instruction, and if there are problems with the instruction itself, it can discuss it with you and then work with you."
Yang Zhilin also mentioned that from an ecological perspective, there are many new opportunities for combining hardware and software. “Mobile phones are the best hardware carrier now, but with the development of AI, will there be better hardware in the future? Will more modalities be incorporated into it? "
In the process of starting a business, Yang Zhilin learned an important lesson: for start-up companies, the organizational method must be able to adapt to new production methods. New technologies often affect production methods to a great extent.
"The development method of large models is very different from that of the Internet. The Internet is more of a planned development. A goal is set, and everything to be developed in the next step is orderly. It is rare to encounter a technical bottleneck that leads to this Things cannot be developed. But large models are emergent. For example, at the beginning of the industry, we have achieved the operation of 10 to the 25th power. Next, we have to perform operations to the 26th, 27th, and 28th power, but we don’t know. What new capabilities will appear at which node, so we need to pay more attention to basic capabilities and let the models continue to emerge." He said that the corresponding change in the production method is not to plant a tree in one place, but to directly plant it. The entire forest was contracted.
Another important insight he gained about entrepreneurship is to "climb the stairs, not just look at the scenery", that is, to strike a balance between commercialization and technological development. "Our important goal at this stage is to continuously improve the capabilities of the model, explore the upper limit of intelligence, and bring a better experience to users. This is our first goal, which is to 'climb the stairs'. We hope to be able to do more Prioritize 'climbing the stairs', not just 'seeing the scenery'."
Recently, Dark Side of the Moon's large model product Kimi has added the "Kimi" function, covering multiple scenarios such as office, entertainment, and practical life. "The development of technology can define new problems and new scenarios." Yang Zhilin said that they have been continuing to iterate recently and delve into more professional scenarios and life scenarios. As new users continue to join, they will further improve the user experience, lower the threshold for using the product, and make the product more valuable. “We are very close to our users, and through their feedback we can also know where there is room for improvement.”
Yang Zhilin's entrepreneurial path seems to be smooth sailing, but he said with a smile that setbacks basically occur in units of "days". "Every day, we may encounter R&D problems, recruitment problems, and financing problems. We have overcome many difficulties. The important reason why we have reached this point is that we have an inclusive environment and the support of talents and resources.”
When Yang Zhilin returned to China for the first time in 2019, he came to Shanghai to participate in the World Artificial Intelligence Conference and gained a lot. This time he came to participate in the 50-person forum and heard many fresh ideas. "We are engaged in AI, and we are very encouraged to see the innovation progress in various industries such as medicine, materials, and robotics. We will have many opportunities for cross-disciplinary cooperation in the future."