The AI ​​PC "coming-of-age ceremony" has arrived. Who is swimming naked and who has come ashore?

Smart Things Author | Yunpeng Editor | Mo Ying In the AI ​​era, everyone is actively embracing AI. As a content creator, I have long been faced with the needs of massive data sorting, data information tracing, and high-quality content generation. In the past year, I have tried a number of "AI PCs" with AI as the selling point. The actual experience is disappointing: AI cannot quickly identify and process the PPT or report being read on the screen, and still requires tedious operations before throwing it to the AI; every question to the AI ​​is like facing a new assistant with "amnesia", and the communication cost is extremely high. Unsatisfactory AI PC experience is not an isolated case. A lawyer friend told me that his use of AI to improve efficiency only increased his worries: he was worried about transferring contracts to large cloud models and was worried about data security. AI’s large amount of “meaningless” output could not be used in court defense, and AI turned a blind eye to a large amount of local case data. A friend who works in industrial design complained to me that AI knows astronomy and geography, but once it reaches its professional field, various professional software becomes AI’s blind spot. AI can instantly generate beautiful pictures, but it cannot help him model in 3D software. Looking at the industry, major manufacturers are intensively launching "AI PCs", but the level of intelligence and AI experience are uneven. A real "assistant" will remember your habits and preferences and understand you better the more you use it, but most "AI PCs" cannot achieve this. At the same time, with the advent of the Agent era, AI has moved from "generative" to "executive" and truly started to "work." However, there are only a handful of "AI PCs" that can truly execute independently and do the work efficiently. The big model has a lot of skills, but he can't really know how to do things efficiently on the PC. The actual problem is very clear: "PC intelligence" is by no means simply equal to "connecting to a large model". Without device-side perception and memory, AI services can only be "rootless water" with weak perception of efficiency improvement; without learning capabilities, AI will never become an existence that "gets to know you better the more you use it." Under such a current situation, it is very difficult for ordinary consumers to choose an "AI PC" in the true sense. If they are not careful, they will be confused by certain marketing terms.