On the afternoon of July 21, it was announced that during the 2026 World Artificial Intelligence Conference, Guo Quanwei, chief scientist of linguistic science, was invited to participate in a roundtable discussion on the theme “What AI is rewriting”. He proposed two core judgments: the next stop for AI is a local model and a local agent; measuring whether an AI product is actually successful is not whether users pay, but whether the company is willing to change the workflow for it. Over the past two years, the AI industry has almost been competing for whose model is bigger and stronger. However, Guo Quanwei believes that what really determines whether AI can enter the core scenario of an enterprise is not the model parameters, but whether the data can stay local. Compared to relying entirely on cloud inference, the local model allows data to be understood, inferred, and executed in the original environment. While ensuring privacy and security, it greatly reduces long-term access costs, and is more suitable for continuous processing of massive enterprise data.

On the afternoon of July 21, it was announced that during the 2026 World Artificial Intelligence Conference, Guo Quanwei, chief scientist of linguistic science, was invited to participate in a roundtable discussion on the theme “What AI is rewriting”. He proposed two core judgments: the next stop for AI is a local model and a local agent; measuring whether an AI product is actually successful is not whether users pay, but whether the company is willing to change the workflow for it. Over the past two years, the AI industry has almost been competing for whose model is bigger and stronger. However, Guo Quanwei believes that what really determines whether AI can enter the core scenario of an enterprise is not the model parameters, but whether the data can stay local. Compared to relying entirely on cloud inference, the local model allows data to be understood, inferred, and executed in the original environment. While ensuring privacy and security, it greatly reduces long-term access costs, and is more suitable for continuous processing of massive enterprise data.