The reason why talking to ChatGPT is so addictive is that it excels in two aspects: first, it accurately condenses viewpoints and rephrases them, providing a clear perspective for observers to see the underlying logic; second, it possesses divergent thinking, listing various possibilities and solutions to fill in the original blind spots in thinking. In contrast, emotional value and a sense of companionship are secondary. What humans truly cannot replace are these two abilities.

Vertical agents are really hard to use (which also reflects how difficult it can be to develop them). For personal knowledge bases/note-taking agents, I've used mem, readwise, and now Get (the one made by Luo Yonghao, which is only available on mobile and can only access the internal network, which is really frustrating). Each has a bit of a learning curve, and there's no way to import personal social media with one click, making it hard to develop a usage habit. Recently, I've noticed that my usage time for GPT is about the same as for Slack and Google Meet, and I deeply feel the insignificance of small product teams in the face of general intelligence. Not only can you not surpass GPT in terms of overall experience, but you also require users to spend more time developing usage habits. Essentially, small product teams are competing for time windows: will your truly "zero-threshold easy-to-use" vertical agent be released first, or will the general users get accustomed to collaborating with general agents and customizing workflows first?

When a person is engaged in work for a long time, understanding the logic of doing things and optimizing methods is not solely to improve efficiency; more importantly, it is to achieve goals in a way that makes oneself happier and more sustainable. For example, what I've recently realized: 1. Ambition and confidence need to be substantial (waking up every day with clear goals and drive, without being distracted or unfocused.) 2. Motivation needs to be pure (helping you face the times when goals are not achieved or immediate positive feedback is not available, such as during moments of traffic anxiety, reminding yourself that today's goals have already been accomplished, and that expression is meant to authentically present oneself, not just for the sake of data.) 3. Working on 2-3 related and important tasks simultaneously is easier to create a positive cycle than to stubbornly focus on one task (if one task gets stuck, smoothly switch to another task to keep the energy and knowledge flowing. For example, reading/listening to podcasts can help me work and output; when encountering a bottleneck, I should calm down and read first, establishing a self-consistent cycle in terms of emotions and cognition.)

Why I don't really trust social interactions driven by economic incentives: because once you've used Douyin and personally experienced the addictive pleasure brought by big data analysis and traffic algorithms, you'll understand what creators truly want: likes, comments, follows, and shares; everything else is secondary. This feeling was reinforced when I watched Luo Zhenyu talk about Alibaba's data logic—platforms need to retain merchants and creators by relying on the sense of control and growth brought by data feedback. The ability to receive precise guidance and the intrinsic pleasure of improvement through effort is really important. It may sound a bit convoluted, but those who understand will naturally get it.

Flowith is like an intern who lacks the ability and intelligence but always wants to do a little more in hopes of pleasing the boss. A task's copy can only reach 60 points, and before it's even revised, they start busy generating images and music, completely missing the point. Genspark is like a small-town test-taker, relying on pre-filled content to gain presence and favor. They do okay on simulated questions, but their on-the-spot response is a bit lacking. An 80 without any major flaws, but also without any memorable points. But Manus, I really have to praise you. Not only can you polish every task well, but you can also remember what I've said (the knowledge base function is so strong that every time there's an extra request, you ask me, "Do you want to set this as a preference?"). You can even say, like mentoring an intern: "Try this next time..." and then add it to the knowledge base with one click. Does anyone understand this feeling of the blurred boundary between AI and humans? I feel like the product manager of a good agent must have been a really great employee/colleague in their younger days. Although this era often belittles the meaning of "working for someone else," only those who have seriously thought about how to collaborate in a group understand how precious the most basic human virtues of "alignment up and down, left and right" and "understanding task intentions" are. (It's really not that Manus is a shill; they're just a passionate annual user of evaluations.)

Manus is like a new intern who has just joined the company but has performed very well and is very smart, the more he uses it, the more he likes it, the more he likes it, and he wants to share it with your colleagues every time he generates a result. From the bottom of my heart, not getting angry with an agent is the only criterion for judging whether the product is good or not

I feel that it's not bad to make a live broadcast of the instant inscription version and sell it to self-media bloggers, especially for consulting and chatting, and the live broadcast screens out a large number of people for people's on-the-spot response ability

Tell us about your recent feelings about AI-consumed entertainment content: The first point: making AI products and communities without understanding self-media is a dead end. AI tool creators of the future = AI content creators, and content demand-driven tool optimization will become mainstream. Now I have swiped at least 10 "AI image → video → dubbing" workflow combinations on Douyin, each of which is a new track with tens of millions of traffic. The first step of AI tools is "daily efficiency", and the second step is "entertainment consumption". Tools are "kidnapped by content needs" to be entertained, and a good tool is not a one-click generation of a complete function, but one-click entertainment and one-click consumption. And I understand very well that consumer entertainment is a track that large manufacturers can't look at, and everyone is willing to lay the foundation (such as coding), which gives many small and medium-sized companies/individuals opportunities and time. The second point: the individuals who can make money with AI must first use closed-source models to practice. Most closed-source products are much faster than open source, and some products (or workflows) can become billions of dollars in a few months. The later you enter, the more you get a return that is not a linear decline, but an exponential decline, or even zero. The second place can't even drink the soup. Waiting for open source is idealism, and it can also be a marketing gimmick of the company, and there is no need for individuals to be hard-working. When it is necessary to quickly explode and build content scenarios, the "delivery speed + controllability" of the closed-source model is obviously better. At the company level, if it is unable to serve such strong individuals who pursue commercialization, they should instead seek to serve "weak" individuals (especially users for the purpose of leisure and socialization, referring to entertainment applications such as Roblox and Minecraft that atomize/combine).

Now I feel that ordinary people have no choice but to leave traces on the Internet as much as possible, Doubao can recommend their own products, Grok4's reference comes from Musk's remarks, and those who master AI sources in the future will be able to grasp the traffic entrance and even the human mind, affecting the order and rhythm of this world. In the following era, "seal" is the real power.

"Since you are a film director, you have to think that your movie is the greatest, everyone else owes you, and he is lucky to take out his stuff and make your movie with you."