On May 28, 2023, Innovation Works Chairman and CEO Li Kaifu gave a keynote speech at the parallel forum “Artificial Intelligence Large-scale Model Development Forum” of the Zhongguancun Forum on May 28, He explained the new opportunities from AI 1.0 to AI 2.0. . Despite the imperfect nature of the large-scale models, including examples of ‘hallucinations’ during response generation, Lee believes these models still have enormous commercial value. He expects the biggest applications of the future to incorporate large-scale AI models, and that “AI-first” companies will emerge as favorites of the times.
Even imperfect large-scale models deliver trillions of dollars of commercial value
Lee pointed out that large models will reshape the product ecosystem and present great opportunities. The main problem with large models at the moment is that they tend to “run wild” during response generation. Solving this problem is difficult because the inference power of large models comes from the same source. Reducing this “exploding” ability to near zero risks compromising the reasoning power of these models. However, there are potential solutions such as data tuning, early warnings, corrections, or transforming large models into smaller models in certain areas to reduce the likelihood of AI “exploding”.
Even a large, imperfect model could potentially have huge commercial value, in the trillions of dollars, according to Lee. “When journalists and lawyers use AI to assist in the drafting or drafting of legal contracts, it cannot be considered a final application because of the potential for errors, and the liability is limited.” It’s still human, but in graphic design and entertainment, especially in entertainment applications, users don’t care about minor errors, like a game character with a slightly longer or shorter beard, or an incorrect line of dialogue. The content of the game is fine because it’s made up, so we can tolerate these imperfect large-scale models in many areas,” Lee said.
Lee gave examples of future possibilities, such as creating games that children in the community want to play, using text to introduce the game, and completing the creation of the game in seconds. In e-commerce and advertising, AI can adjust advertisements and images for each product based on each individual’s needs, cognition, education level, and buying habits, increasing purchasing power. Of course, there are also regulatory issues. Legal supervision is required if the written content is false or harmful to users. But these two he examples show that large-scale models are more than just question-and-answer engines. Transform the ecosystem of apps across every industry.
Models at Scale: A Historic Opportunity You Can’t Miss to Redefine Every Application
Lee believes that large-scale models will not only revolutionize artificial intelligence, but will also create significant platform differences between companies. What stands out are the applications with “AI-first” applications. He explained that “AI first” refers to companies that wouldn’t exist without large-scale AI models.
Lee gave the example of the mobile internet boom, including apps like Meituan., Didi and Douyin maximized the capabilities of mobile phones to achieve explosive growth in product applications. Therefore, if you want to start a business or invest in mobile internet, you should choose apps that cannot work without mobile internet. Similarly, if you develop an app or start a business in today’s AI space, you must choose an app that cannot function without a large-scale AI model. This is what “AI first” means, and apps like this will become mainstream in the future.
See also: Meituan’s AI model development predicts big changes
Lee believes that the biggest applications of the future will not function without large-scale AI models. In his view, large-scale AI models present a historic opportunity that cannot be overlooked. These have become the greatest platforms of all time, ten times more important than Windows and Android, rewriting all applications, restructuring human work, amplifying human intelligence ten times more than his, and many replace repetitive tasks.
But Lee concedes that the technology has a long way to go in its application and is unlikely to see real-world use cases in the next three to five years. “In the development of large-scale models, China faces several challenges, such as a slower start compared to the United States, lower computing power, and less experience with large-scale models. However, the Chinese market is vast. And the economy is strong, and the Chinese government is poised to deal with this job restructuring, effectively moving more Chinese into cooperative positions, which is difficult in the West. Yes, China has a huge number of AI scientists and has a great advantage in terms of human resources.I believe that with the cooperation of the government, large enterprises, start-ups and investment companies, we can quickly overcome these challenges.” said Lee.
