In my opinion, the difference between smart and less smart AI models often comes from the differences in datasets. Just like when I compared the usability of answers to local life issues between Tencent Yuanbao and Deepseek, I found that although Tencent Yuanbao's core is still Deepseek, its answers are much 'smarter' than those of Deepseek itself, allowing for direct usability based on the answers.

At its essence, this is because Tencent Yuanbao can directly access a large amount of data from WeChat public accounts, which is not completely open. Within this database, there is a wealth of experiences and viewpoints shared by self-media. It’s easy to imagine that if Xiaohongshu could create an AI, it might be even more impressive in life experiences than Tencent Yuanbao.

This issue proves the importance of high-quality data. While AI can help people find good restaurants and their contact information, only humans can create and innovate restaurants; creativity is still something AI cannot achieve.

Moreover, the recent report from Tiger Research mentions the crisis in the data field. Due to the proliferation of AI content, high-quality data resources may face depletion, which poses significant challenges for data-driven AI models. Even more troubling is that much user-generated content is used for AI training without permission, and original authors often do not receive recognition or financial compensation.

Many teachers are saying that @campnetworkxyz is about to issue tokens, and I've seen quite a few updates related to the Camp ecosystem recently, feeling like a new version of $IP.