AI startups continue to be valued at $1 billion light-years away seeking financing

The wave of entrepreneurship ignited by ChatGPT has attracted a group of entrants. "LatePost" learned that the new projects that are in the preparation or financing stage recently include a new company with Shen Xiangyang, chairman of IDEA Research Institute, as an advisor, the re-entrepreneurship of Yang Zhilin, co-founder of Circular Intelligence, the new company of Zhang Qianchuan, former head of user products of Toutiao, a subsidiary of ByteDance, and a new project of Wang Meihong, former head of international technology of Kuaishou.

In addition to the new company Lightyear, Meituan co-founder Wang Huiwen, which started earlier, has recently launched a second round of financing, with a pre-investment valuation of about $1 billion. It is understood that Kuaishou Chairman Su Hua is also paying close attention to the direction of large models and has invested in several projects.

The large model industry chain, from the bottom to the top, can be divided into the basic layer that provides computing power, the framework layer that provides training tools, the model layer that provides the large model itself, and the application layer based on the large model.

The new projects described above focus on the model layer and the application layer. One of the model-layer startups is a new company founded by Zhang Jiaxing with Shen Xiangyang as an advisor. Zhang Jiaxing is the head of the Cognitive Computing and Natural Language Research Center of the Guangdong-Hong Kong-Macao Greater Bay Area Institute of Digital Economy (IDEA Institute), and before joining IDEA Research in 2021, he was a researcher at Microsoft Research Asia, a senior algorithm expert at Ant Group, and the chief scientist of 360 Mathematics. He has talked about one of the most important changes that ChatGPT has brought about by changing the way AI capabilities reach end users.

Shen Xiangyang served as Microsoft EVP (Executive Vice President), he worked at Microsoft for 23 years, participated in the founding of Microsoft Research Asia, responsible for the development of Microsoft's Bing search engine and intelligent chatbot Microsoft Xiaoice. In 2020, he left Microsoft to become Chairman of the IDEA Institute.

Another entrepreneur who will invest in model-level entrepreneurship is Yang Zhilin, co-founder of Recurrent AI. His new company has received investment from institutions such as Sequoia China and GSR Ventures.

Yang received his Ph.D. in Computer Science from Tsinghua University and received his Ph.D. in Computer Science from Carnegie Mellon University in 2019, where he worked at the Facebook Artificial Intelligence Institute and the Google Brain Institute. On the basis of the Transformer architecture born in 2017, Yang Zhilin later proposed Transformer-XL, XLNet and other achievements as the first author, which were recognized in the field of large language models. Founded in 2016, Circular Intelligence is dedicated to helping companies improve sales with artificial intelligence technology. Yang has been involved in circular intelligence work during his studies, and has been the company's head of AI and products since 2019, around the time of his Ph.D. graduation.

Entrepreneurs who plan to apply based on large models include Wang Meihong, former head of international technology at Kuaishou, and Zhang Qianchuan, former head of user products at Toutiao, a subsidiary of ByteDance.

It is understood that Wang Meihong plans to make a video version of Midjourney. The latter is a Wen Sheng Tu app that earned more than $100 million last year. Wang wants to go one step further and let users enter text to generate videos. Meihong Wang, who worked on Facebook's Newsfeed product, joined Kuaishou in 2021 and has recently left. Zhang Qianchuan, who plans to make a social application in the AI era, worked at Baidu Search and Zhihu, joined ByteDance in 2016 to be in charge of Toutiao's user product team, and has been away from Byte for a while.

An investor who has been focusing on the technology field for a long time said that startups can choose to develop large models like OpenAI, which requires cleaning data from scratch, purchasing large computing resources, and recruiting senior talent; It can also be fine-tuned based on open source large models, which requires fewer resources, and the difficulty lies in selecting more accurate and professional data so that the model has differentiated capabilities. There is a higher threshold in both cases. Large domestic companies have successively released large models since 2019, and this year are mobilizing more forces to make large models, leaving a short window of time for entrepreneurial teams. Just look at the quantity, more items will appear in the application layer.

However, a group of entrepreneurs with the most technical strength, strategic ambition or resource integration ability have chosen the high-risk, high-return model layer direction that may bring about big changes.

According to "LatePost", in addition to the light years founded by Wang Huiwen, the new company of Wang Xiaochuan, the former CEO of Sogou, and the new business that AI company will open by Mobvoi founder Li Zhifei have chosen to make both large models and model-based applications. The model layer provides technical support, and the application layer provides data feedback to accelerate technology and business evolution collaboratively.

Similar to Wang Huiwen and Wang Xiaochuan, some of the previous batch of successful Internet entrepreneurs are also considering investing in a new round of artificial intelligence boom. Kuaishou Chairman Su Hua has recently invested in several generative AI-related companies. A person close to Suhua said he has been looking at opportunities that could lead to huge innovations, such as new energy and artificial intelligence.

Yang Luyu, founder of Musical.ly and former head of education hardware at ByteDance, is also considering investment. He told LatePost that he might invest in some good teams at the application level, especially in the to C area where he can help.

A technology investor said that some institutions are still waiting for application-layer companies, especially those facing the domestic market, and although OpenAI has not banned Chinese companies from using its APIs, it is uncertain whether Chinese customers and users can continue to use foreign large model capabilities in the future. When China's own big model is more mature, the number of application-layer startups facing the domestic market will be greater. (Wen 丨 Shi Xian Editor 丨 Cheng Manqi)