In a dialogue at the Future Science Awards on November 8, Zhang Yaqin asked Yao Shunyu, Tencent’s chief AI scientist: Where is the moat of AI companies?
Yao Shunyu's answer is the product layer, not the model layer. What is really difficult to replace are those products that understand the user background, hold proprietary data, and have gained the trust of users.
He took WeChat as an example. If AI is embedded in WeChat, it knows your schedule and your social relationships. Even if the opponent's model is smarter, it will not be able to provide the same service.
Zhang Yaqin joked: According to this logic, Anthropic may not be worth two trillion US dollars in the future.
There is a serious question behind this joke. In the past two years, the industry's default model capability has been a barrier, and whoever scores higher is worth more. However, model capabilities are converging rapidly. Open source models can catch up with closed source models in a few months, and barriers are migrating to data and scenarios. Tencent happens to have the biggest scene in its hands.
Of course, there are counterexamples to this judgment. If the next generation model widens the generation gap in capabilities, users may be willing to migrate for better results. The product moat can only block opponents who are "almost easy to use", but cannot block opponents who are "much easier to use".
On the same occasion, Yao Shunyu also talked about risks: He believes that AI will soon master biology, and related security risks may be more difficult to control than cyber attacks.