The “inspection standards” for AI implementation tend to be diverse

China News Service, Shanghai, October 17 (Gao Zhimiao) The Global Wealth Management Forum·2026 Shanghai Suhewan Conference will be held from the 16th to the 17th. At the "AI - From Technology Frontier to Industrial Application" roundtable forum held on the 17th, the guests' discussion did not stop at "what AI can do", but repeatedly asked a more fundamental question: What is the standard for judging the progress of the AI ​​industry?

Song Haitao, president of the Shanghai Institute of Artificial Intelligence and a member of the United Nations Independent International Scientific Group on Artificial Intelligence, summarized the changes in AI in the past year with four key words: "acceleration, knowledge equality, globalization, and appropriateness", and pointed out that the inspection method itself is also changing. “In the past, we still needed to prove ourselves through lists, but now a large number of markets and customers have actually learned to use models to evaluate our solutions and technical capabilities.” In his view, this shift from “self-certification” to “the market voting with its feet” is itself a signal that AI is moving toward industrialization.

Li Jiaqing, President of Legend Capital, proposed another test criterion from the perspective of industry and investment. He believes that discussing "whether large models can make money" is a false proposition, because large models will run through the entire industry and are a must-have element in the end-to-end industrial chain. It is meaningless to talk about enterprise applications or infrastructure without this element; the industrial structure itself is evolving in layers. "If you make more money this year, your share of the money will be handed over to me tomorrow." All parties are dynamically adjusting and extending downstream in the layered structure.

Faced with the question of "will everyone in All in be too crowded?" Li Jiaqing advocated covering chips, infrastructure, large models, and industrial applications in an investment portfolio, and staying with them for a long time with a "cultivation" mentality. He specifically reminded that non-market factors will cause long-term logic to be short-term, causing things that were originally not volatile to fluctuate violently. Therefore, in addition to "long-term capital" and "patient capital", we must also emphasize "seeking truth and being pragmatic", and do not regard long-distance running as short-distance running.

Li Dahai, co-founder, chairman and CEO of Wall-face Intelligence, describes the current stage as a "technological explosion": the knowledge density of large models doubles approximately every 100 days, client-side models jump from 1B to 35B within a year and can be run on mobile phones, and AI begins to participate in training AI and solving mathematical problems that humans have not yet solved. He believes that model capabilities "diverge and converge at the same time." The long-term value of technology is easily underestimated and short-term expectations are easily overestimated, which is the source of industry anxiety. “The growth that large models bring to the entire society is highly deterministic and accelerating. I think it is a process that continues to happen.”

AI industrialization has moved from the "storytelling" stage to the stage of testable value realization, and the inspection standards for implementation are not a single dimension, but the result of mutual calibration of multiple scales.