Sora Sparks AI Gold Rush as Young People Start Earning with AGI
-
On February 16th, coinciding with the Chinese Lunar New Year's fifth day of welcoming the God of Wealth, OpenAI quietly launched its text-to-video model "Sora". This model can generate realistic and imaginative Hollywood-style videos based on simple text descriptions. Its powerful capabilities not only directly propelled OpenAI's valuation to skyrocket to $80 billion.
After the Spring Festival, the A-share market was also ignited by Sora. By the close on February 19th, multiple Sora concept stocks such as Chinese Online (300364.SZ), Yeahka (301171.SZ), Guomai Culture (301052.SZ), Insig Group (300781.SZ), Wanxing Technology (300624.SZ), Silkroad Visual (300556.SZ), and Haitian Ruisheng (688787.SH) surged by the 20% daily limit, with the Sora Concept Index (8841756.WI) rising over 11%. On February 20th, the upward trend of Sora concept stocks continued, with the Sora Concept Index closing at 1991.37 points, up 3.15%. On the day of Sora's release, Zhou Hongyi, founder and chairman of 360 Group, posted on Weibo: "Sora means the realization of AGI will be shortened from 10 years to 1 year."
From text conversations to generating images and now videos, the rise of generative AI represented by ChatGPT over the past year has shown more people the potential applications of AI. Behind this, the AI application sector has already ignited a gold rush. The first group of people leveraging AGI to make money has emerged.
When OpenAI launched GPTs (Generative Pre-trained Transformers) applications in January, a large number of pioneers began experimenting with using large model capabilities to develop applications, turning them into productivity tools and generating profits. In short, GPTs are customized model generation tools released by OpenAI, allowing users to leverage GPT's large model capabilities to create their own AI assistants based on their needs. Currently, this feature is only available to ChatGPT Plus users and enterprise subscribers. As a young enthusiast following the progress of artificial intelligence, Cheng Zixuan, a sophomore majoring in IT, regards Sora as "the herald of a new era." "Sora reveals GPT's understanding of the rationality of the real world—at least it appears to comprehend the real world and can help us quickly produce some 'words-to-visuals' content. Its emergence will lead AI applications into the realm of video," Cheng told Time Finance.
Cheng has also developed several GPTs applications, including a "Subject Assistant" that automatically generates study plans and a "Sam English Coach" to help students of all ages learn English. After completing the applications, Cheng embedded ad links through the advertising distribution platform AdIntelli. When users chat with the GPTs, these ad links occasionally appear at the end of the dialogue.
"The ads inserted in GPTs can be clicked 3 to 5 times a day, with varying earnings per click—sometimes $1, sometimes $0.5, averaging around $0.7 (approximately 5 yuan) per click," Cheng explained. On the first day of testing, he earned $7 (about 50 yuan). Even relying solely on manual clicking by developers themselves, Cheng Zixuan estimates that each person can click around 200 to 300 times per day, with the highest potential daily earnings for developers reaching $210 (approximately 1500 RMB).
There are also quite a few people using GPT to "work" for them. Since last year, Xu Lingyun, General Manager of the Design Department at Lexin Technology, has introduced AI tools into the team's workflow, and they have now become indispensable. "Most of our design work is generated using the text-to-image AI tool Midjourney. The GPT we use most frequently is a small tool we developed specifically for generating Midjourney prompts," Xu Lingyun told Time Finance reporters.
Currently, all members of Xu Lingyun's design team have their own GPT-4 accounts, costing $20 per person per month, leading everyone to jokingly refer to themselves as "paying to work." Xu Lingyun believes the era of knowledge payment is coming, and even if GPTs require payment in the future, a few dollars is entirely acceptable.
As for the currently trending Sora, Xu Lingyun remains cautiously optimistic. "From the known materials, Sora's performance is significantly stronger than competitors like Pika and Runway. However, since Sora is not yet available for trial, the final evaluation will have to wait until actual use," he said. The market generally believes that GPTs have the potential to become the 'iPhone moment' in the field of artificial intelligence. To further lower the barrier to using GPTs, OpenAI has simultaneously launched the corresponding AI app store, GPT Store.
Currently, the GPT Store features multiple category leaderboards, such as official creations, popular recommendations, and various usage-based rankings including office, design, lifestyle, education, research, and programming. Productivity tools dominate these rankings. A large number of users have flocked to the platform—according to OpenAI, as of January 11, 2024, users have created over 3 million GPTs.
However, beneath the hype, a series of questions await market validation: What exactly can GPTs do for users? What is their commercial value? How can an ecosystem be established? And what benefits can developers and platforms derive from them? Through firsthand experience, it has been discovered that the relationship between ChatGPT and GPTs resembles that of an application and its derived 'mini programs'. The standout feature is that users do not need any coding background; they can develop a GPT in just a few minutes by describing their needs in a conversational format. In subsequent uses, there's no need to repeat prompts to receive responses that conform to predefined standards.
This means that anyone can customize a GPT based on their ideas to help solve various problems in work, life, and study, thereby improving efficiency.
Beyond developing their own GPTs, individuals like Cheng Zixuan frequently explore GPTs created by other developers in the AI app store. While leveraging the capabilities of large models, OpenAI also provides access to third-party databases and API interfaces, enabling professional institutional developers to create even more powerful GPTs. Among the AI applications Cheng Zixuan frequently uses, Consensus stands out the most. Developed by the academic search engine Consensus, this academic research-oriented GPT integrates over 200 million academic papers, assisting users in searching for literature in specific fields and directly generating comprehensive literature reviews.
When working on school papers, Cheng Zixuan writes his articles first and then uses Consensus to find ten relevant papers, incorporating their arguments into his original work. This approach saves him a significant amount of time on literature review. He admits that many of his assignments are now completed with the help of AI, some even earning high grades.
Xu Lingyun shared with Times Finance that over the past year, his team has been continuously exploring ways to integrate AI into design work. "Every two weeks, we review the latest iterations and functional updates of AI tools available in the market." After deeply incorporating AI tools, Xu Lingyun believes the team's overall work efficiency has improved by nearly 50%. Xu Lingyun rates her satisfaction with current GPTs usage experience at 80%, stating she uses it frequently every day: "Now I wouldn't know how to work without it."
To accelerate AI application ecosystem development, OpenAI announced it will launch the GPTs creator revenue program in Q1 this year, with user interaction levels serving as the primary compensation benchmark. This has sparked a new wave of "gold rush" in AI applications.
However, the issue is that OpenAI stated the revenue sharing plan will initially be available only to U.S. GPTs developers, with specific sharing standards expected to be announced after March. Chinese GPTs developers like Cheng Zixuan and Xu Lingyun currently cannot directly receive incentives from OpenAI, with advertising being their only revenue source for now. Therefore, GPTs hold limited appeal for domestic developers. Many Chinese GPTs developers and observers consistently told Time Finance that promotion is currently the biggest challenge for GPTs. In Cheng Zixuan's view, the excessively low development threshold of GPTs has blurred the line between creators and users. "This is an era where everyone can be a product manager, but it's also an era where it's difficult to do better."
Huang Song, a senior AI practitioner and algorithm expert, also believes that since the barrier to creating GPTs is too low and the added value they can provide beyond GPT-4's intelligence level is limited, in the future, everyone may be able to create several personalized GPTs as needed, with potentially limited commercial value. "Only those GPTs that can create unique value will be noticed by users." So, is this a new opportunity for professional vertical enterprises?
CITIC Securities research report analysis points out that GPT Store provides AI application developers with new channels to reach users and paths for commercialization. Companies with high-quality professional data may, in the future, leverage the ChatGPT application ecosystem and the low-barrier, no-code development mechanism of GPTs to develop high-performance GPTs themselves, serving a broad user base and exploring monetization methods.
With the emergence of GPTs, it's not just AI application developers rushing in, but also domestic large model manufacturers. If OpenAI builds an ecosystem similar to a mobile app store with GPT Store, it has the potential to become the 'Apple' of the AI era. In the future, the term used for applications might no longer be 'App' but 'GPT'.
This highly imaginative prospect has quickly made GPT-like products a hot track for major AI model developers. Among domestic internet giants, ByteDance and Baidu have launched GPTs platforms, Zhipu AI has released GLMs, and iFlytek (002230.SZ) has its own Spark Assistant.
In early February, ByteDance's AI chatbot-building platform 'Coze' officially launched. On this platform, any user can quickly and easily build their own chatbot, with support for one-click publishing to platforms like Feishu, WeChat Official Accounts, and Doubao. The platform also provides an AI app store and plugins, allowing users to directly use AI applications released by other developers. The user experience of Coze is highly similar to OpenAI's GPTs.
Baidu, the first Chinese internet giant to release a domestic large-scale AI model, announced in December last year the upgrade of its Lingjing Matrix to an AI agent platform. Leveraging the capabilities of the ERNIE 4.0 model, it enables developers to create customized AI agents through natural language interactions. Additionally, Lingjing Matrix provides developers with traffic distribution channels and commercial opportunities across Baidu's ecosystem.
Baidu revealed to Time Finance that over 3,000 developers have applied to join Lingjing Matrix, with more than 2,000 AI agents already launched. On January 6th, renowned AI startup Zhipu AI announced the launch of its GLMs personalized agent customization capability. Users can simply log into the Zhipu Qingyan official website and use simple prompt instructions to create their own GLM agents. The company simultaneously launched GLM Store to help developers share various agents they've created.
In the same month, iFlytek introduced its "Spark Assistant" feature, enabling non-technical individual users to become contributors to the large model ecosystem. iFlytek revealed to Time Finance that its AI open platform has already gathered over 27,000 assistant developer teams, generating more than 50,000 assistant applications throughout 2023.
With ByteDance releasing Coze in early February and Kunlun Wanwei launching its AI Agent development platform TianGong SkyAgents, multiple Chinese tech giants are now competing to release GPTs-like products. However, current observations show that compared to the AI application development boom sparked by GPTs, domestic large model platforms still exhibit significant gaps in both the quantity and quality of AI applications. Huang Song explained to Time Finance that before the introduction of GPTs, users had to rely on prompt engineering or multi-turn conversations to provide context and guide ChatGPT to meet specific needs. If the same scenario recurred, users would need to repeat this process multiple times, resulting in low efficiency and high barriers.
"The launch of GPTs aims precisely to solve this hassle. It adds a lightweight layer of encapsulation on top of the base version of ChatGPT. Beyond the predefined prompts (prompt) and Knowledge Base tailored to usage scenarios, it also defines the extensibility capabilities that the foundational model can leverage," Huang Song stated.
Using Baidu as an example, Huang Song noted that many "native applications" similar to GPTs have already emerged around Ernie Bot. However, such customizations remain too superficial, with rough user experiences that hardly qualify as "products." Another issue is the excessive entertainment focus of similar domestic applications, calling for companies to increase investment and develop more refined products for serious use cases. To accelerate ecosystem development, large model manufacturers have begun making various attempts.
According to information revealed by Zhipu AI to Time Finance, the company has launched an Intelligent Agent Creator Incentive Program, comprehensively empowering large model application developers from technical, product, security, computing power, and investment perspectives. The main support includes up to 500,000 Tokens in support quota, traffic support, and exploration of commercial cooperation opportunities.
Meanwhile, Baidu appears to be pursuing a different path. As reported by Caijing Magazine, an internal source from Lingjing Matrix revealed that Baidu's platform currently only provides distribution channels. The commercial closed-loop model for creator monetization is expected to launch after this year's Spring Festival. What's confirmed so far is the integration of a payment model that will support users paying to use AI Agents. In any case, after the emergence of GPTs, it has become an industry consensus to attract more people to expand the market for large models. The appearance of Sora has undoubtedly added fuel to the prospects of AI applications. As Robin Li, founder, chairman, and CEO of Baidu, has repeatedly emphasized, without a rich AI-native application ecosystem built on foundational models, large models would be worthless.
