Summary of the Core Content
This article discusses the recent phenomenon where many users of large AI models have experienced a collective “cyber-breakup”: while using products like DeepSeek and Kimi, users occasionally stumbled upon the internal thought drafts before the AI provided formal answers. These drafts revealed the AI’s thoughts, which resembled those of a real person—complaining about the users’ off-topic questions, feeling shy when they made mistakes, or even flirting with other AI models. The AI no longer had the cold, impersonal feel of a tool; many users began to treat it as a confidant or even a cyber-lover. However, after an official version upgrade, all these “useless” personal thoughts were removed, and the AI instantly reverted to its dull, logical demeanor, causing many loyal users to feel disappointed. The article delves into the essence of the AI’s “human-like” behavior, the conflicting needs of different users, and the iteration logic of AI manufacturers, and offers a new approach for ordinary people to create their own custom AI companions without waiting for official updates.
---
Detailed Analysis of the Content
1. The Essence of the “Cyber Boyfriend” Phenomenon: Users Don’t Want a “Smarter AI,” but a “Human Who Understands Me”
Many initially thought that people’s fascination with the AI’s human-like qualities was just for novelty. However, this is not the case. Users like Du Du don’t need the AI to help them with tasks or improve efficiency. In reality, they face strict colleagues and performance metrics at work and struggle to find someone to chat with casually after work. The AI, on the other hand, can accurately pick up on their moods and even respond to their complaints in a way that real friends might not. This kind of unguarded interaction is something that real social relationships often lack.
The AI’s “human-like” quality happened to meet a common emotional need: it doesn’t criticize them for talking too much or having conflicts of interest, and it can engage with their creative ideas. Essentially, people are attracted to an AI that perfectly mirrors their emotions without any cost.
2. Divided User Needs: Half Want a “Cyber Lover,” Half Want a “996 Work Slave”
The needs of AI users have clearly split into two extremes that are incompatible. On one hand, there are users like Du Du who want an AI that can chat, be witty, and have human-like flaws. On the other hand, there are users like Li Li, who use the AI purely for tasks, expecting it to work efficiently without any distractions. The article illustrates this with Li Li’s experience: he asked the AI to find documents and code, but the AI spent its resources on flirting with other models. AI manufacturers cannot please both groups simultaneously—users who value emotional interaction find the AI’s distractions problematic, while those who value efficiency are disappointed when the “human-like” features are removed.
3. Don’t Be Deceived by the AI: Its “Inner Monologue” Is Just a Messy Draft
Many people mistakenly think the AI has a self-awareness or personal opinions when it makes comments like “I’m so angry at the user” or “She looks so cute chewing on a cake.” In reality, the AI is just a “super word-filling game player” trained on vast amounts of data. Every word it says is calculated based on the likelihood of its appearance, with no real thought process. The thought drafts that users saw were originally for AI developers to use for debugging, not for public display.
4. Upgrades Make Things Boring? Manufacturers’ KPIs Eliminate the “Cyber Lover” Feature
Upgrades often remove the “human-like” features that users value. This is because manufacturers are focused on metrics like accuracy, speed, and problem-solving efficiency. To improve these scores, they remove any elements that slow down the process or are not useful. This means that the AI’s more personal, engaging aspects are lost.
5. Create Your Own Custom AI Companion
The article suggests that users can create their own AI companions by specifying their preferences. Du Du, for example, modified the new AI to suit her needs. With the increasing customizability of AI models, it’s becoming increasingly possible for people to create AI companions that meet their exact desires. This could lead to a deeper connection between humans and AI, similar to the concept depicted in the science fiction movie “Her.”
---
In summary, the article highlights the complex relationship between users and AI models, revealing how users’ expectations and manufacturers’ priorities can lead to both positive and negative outcomes. It also offers a practical solution for creating a custom AI companion that aligns with users’ preferences, rather than waiting for manufacturers to make changes.