Source: Future Technology Power

Author: Xue Liang Neil

Original Title: "Staying Firmly with Wave of Big Models"

In a sense, the big model industry has entered a "calm" state.

Calm is even a euphemistic way of putting it. Many people would more bluntly say that the big models have "shut down." The launch of the Battle of Big Models brought about a dazzling array of big models, and eager investors and entrepreneurs with top resources took their positions, but then the fervent state slowed down and cooled off.

This forms an interesting and distinct contrast with the situation at the first big model industry forum held by PingWest three months ago. At that time, everyone was eager and optimistic, expressing confidence. People believed that with the speed of China, after the emergence of universal big models, there would be drastic changes in the application layer and the entire industry chain.

Now, three months have passed. Many questions have become interesting. For example, among so many models, which ones can match or come close to the level of GPT-4 and Llama-2? What will be the breakthrough progress in the application layer, and what form will killer applications take, and when will they appear? After training, do we have enough computing resources ready for large-scale inference of big models in our industry chain?

These questions can be summed up as: What stage is China's big model industry currently in, and where is the next direction after the noisy Battle of Big Models?

We want to answer this question with a forum. On August 23, the "Model Reflection: Big Model Industry Ecological Forum: Boiling Capital, Waving AGI," hosted by PingWest and its content brand Future Technology Power, was held in Shanghai. We had in-depth discussions with friends from the industry chain, academia, and investment community on the iPhone moment of big models, the outbreak of vertical big models, the landing application of AIGC industry, and the design of computing power architecture.

Is the iPhone Moment of Chinese AI Big Models Here?

The theme of the first half of the forum was how to understand the revolutionary nature of big models.

The leap in technology and how it translates into practical landing is a question that every player aspiring to this competition needs to consider.

PingWest Founder and CEO Luo Yihang stated in the opening speech "Big Language Models Should Not Be a Dazzling Firework Show," "Big language models should not be a dazzling firework show. From the heat of the first half of the year to the cold of the second half, it should turn to production, turn to landing, rather than being a lively thing in the past. I agree. They say big language models should not be a dazzling firework show. In fact, big language models cannot be said to be a dazzling firework show. This industry is also developing, and the ups and downs we are experiencing now is something we must embrace."

Fu Sheng, Chairman and CEO of Cheetah Mobile and Chairman of Liehu Xingkong delivered a keynote speech titled "Has the Big Model Really Created an iPhone Moment." Fu Sheng pointed out that big models have brought about an interactive revolution and a productivity revolution, and the future ecology depends on the joint efforts of the industry.

Regarding the opportunities for enterprises in the era of big models, Fu Sheng believes, "The competitiveness of an enterprise lies not in human resources, but in intelligent resources. Many startups in Silicon Valley used to hire people, from angel rounds to Series C, or started buying equipment from Series B. Now it's a three-piece set, big models, databases, etc. This will bring about social change, and in the future, the longer people work in front of computers, the simpler their work will become."

Li Zhi, Dean and Chief Analyst of iResearch Wisdom Institute shared a keynote speech "Analysis of the Impact of Big Models on Industrial Digitization." The most important thing in high-quality development is the improvement of total factor productivity, and today we need to rely on technological and data elements to achieve innovation-driven development.

Allen, Chief Architect of AI&HPC Application Software at Inspur Information gave a keynote presentation on "The Way of Computing Power in the Model Era," stating, "Big models are first and foremost a huge engineering challenge, and some model capabilities also come from practice. How to build a ten-thousand-card AI platform is a huge challenge for different vendors or different customers. Large-scale training will also encounter various hardware issues, including in the training process, in network design, etc., which were not encountered in the training process of small models before. GPT-4 brings a lot of engineering experiments, and the closed source hinders the industry's reproduction."

Wu Runze, Technical Director of User Profile Group at NetEase Fuxi stated that the efficient human alignment of the landing application of big models is a work with huge potential, and it also brings disruptive impact to the gaming industry. The intersection of artificial intelligence and games is expected to help the development of artificial intelligence with the massive data and simulation environment of games.

The Last Mile to Landing

In the second half of the forum, we focused more on the landing application side of big models. How to truly train and use big models, and achieve a perfect landing closed loop on the industrial side, is our focus.

Zhang Jiaxing, Chair Scientist of the Cognitive Computing and Natural Language Research Center at IDEA Research Institute shared the latest development status of the "God List" big model and revealed that the Ziya model has strengthened a single capability based on the comprehensive model. He said, "We can form dedicated models, build our own AI products, or apply models in different scenarios. This may be the image of our entire big model industry."

Jian Renxian, Founder and CEO of Bamboo Intelligence viewed the emergence of big models as "the most exciting moment in my 30-plus years of professional career." Bamboo Intelligence is committed to helping enterprises complete the last mile of landing big models: "Big language models have brought about a new paradigm shift in software, which will be the most important work in the future."

In addition to the keynote speeches, we also set up three roundtable forums specifically for the commercialization of big models, focusing on the field applications of vertical big models, the disruptive impact of generative AI on the design industry, and the new opportunities emerging in the AIGC field.

Is Vertical Big Models the Next Outburst Point?

The advantage of vertical big models lies in customization, deep industry know-how background, and relatively small computing power scale. Based on these advantages, what level can current vertical big models reach? What kind of disruptive impact will they have on their respective fields in the future?

Yang Yifan, Vice President of Xinghuan Technology said, "We need to provide capabilities for B-end enterprises from the application layer to the bottom IT system, which is what we are really doing, and it is also what we are doing together with all of you."

Liu Jiangxian, CSO and Chief Strategy Officer of Daguan Data believes that to advance artificial intelligence, three elements need to be addressed: computing power, algorithms, and data. Enterprises need high-quality data while also being careful not to disrupt the intelligence generation of models.

Wang Chen, General Manager of Shanghai Branch of Zhongke Wenge cited vivid examples from his own company, "All international events ultimately have the shadow of Wenge. Our big model covers information from 42 languages and 182 countries, and can greatly improve labor efficiency in the field of news and public opinion through keyword extraction."

Zhu Junjie, Vice President of Hubo Technology also acknowledged the potential of big models: "Almost all intelligent scenarios, such as automobiles, education, and other fields related to intelligence, will be disrupted by big models. As long as it is related to intelligence, it will be reconstructed by big models in the future."

What Can Designers Do with Language Models?

Generative AI first had a significant impact on the visual design industry. Will big models help the design industry explore broader creative boundaries or completely overturn the current design industry landscape? With these questions in mind, PingWest Founder and CEO Luo Yihang, Lingdong AI Founder Lei Haibo, Chen Qi, Founder of Mogujie and WeShop, and senior AI practitioner Chen Yixin engaged in a discussion.

Lei Haibo, Founder of Lingdong AI looked forward to whether the work of future designers will be replaced by AI, and he believes that "human wisdom is still very important when creating new divisions. We in the design circle also believe that big models can replace some basic work, but human beings are still needed for some technical divisions. This is the only value we have as human beings in this world."

Chen Qi, Founder of Mogujie and WeShop, stated that under the influence of AI in the future, visual design will go in two opposite directions: one is "extremely integrated with the real world," serving various daily functions in the real world; the other extreme is to fully extend in the virtual world.

Senior AI Practitioner Chen Yixin believes that the most significant mission of AIGC is to create people's imaginary world—also known as the metaverse—at a very low cost. Creating immersive visual designs is a problem that AIGC can solve.

Entrepreneurial Challenges and Investment Opportunities in AIGC

The final roundtable discussion of the forum was hosted by Wang Zhaoyang, head of the content center at PingWest; Bai Zeren, Vice President of Linear Capital; Wang Zhiwu, Founder and CEO of Yuanjing Technology; Zhong Guangqing, Co-founder and CPO of Fabarta; and Qiu Yiwu, Founder of Zao Wuyun, as dialogue guests, discussing the entrepreneurial challenges and investment opportunities in AIGC.

Bai Zeren, Vice President of Linear Capital, believes that the process of the emergence of vertical applications of big models is relatively slow, which means it is not an "opportunistic" opportunity, but a relatively long-term development process.

Wang Zhiwu, Founder and CEO of Yuanjing Technology, revealed progress in virtual digital humans. "We have made many attempts in the two tracks of virtual and digital humans, creating concepts of intelligent front desks and intelligent reception. The most important mission for us in the second half of this year is how virtual humans can be integrated with the C-end. We hope to create service-oriented virtual humans with memory, soul, emotions, and perception."

Zhong Guangqing, Co-founder and CPO of Fabarta, shared progress in the infrastructure construction of the big model industry chain. How to better integrate big models, data, and business is a pressing pain point for enterprises facing big models. Although there are many big models, there is still a lot of room for improvement in their effectiveness, usability, and accuracy in the landing process. Currently, both academia and industry are focusing on this issue.

Qiu Yiwu, Founder of Zao Wuyun, as a serial entrepreneur, shared his views on the new opportunities in the AIGC industry, which is to turn the technological design and innovation capabilities represented by big models into a platform, providing product innovation capabilities to manufacturing and brand enterprises in the industry chain. This wave of AIGC is particularly advantageous for industries that have not been industrialized, and industries that have traditionally relied more on manual labor, such as jewelry processing, may be the first to enter the AI era.

As a technology content platform and innovation connector with a global perspective, PingWest has always closely followed the development of China's big model industry. At this forum, PingWest also launched the collection of nominations for the 2023 AI Industry List, aimed at major enterprises and investment institutions. The complete list will be officially released at the 2023 PingWest Technology Innovation Conference in November 2023.

Any talk of "cooling down" about big models can be put to rest after this forum. In fact, we prefer to see the "silence" in the current development of big models in China as a sign of its positive development in the deep field. Whether it is the construction of computing power or the deep cultivation in vertical industries, just like training a big model that surpasses GPT, it is destined to be a long-term effort. However, PingWest looks at the development of big models in China with great optimism. We firmly believe that China's artificial intelligence industry will eventually forge its own path of innovation and breakthroughs.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。