Skip to content
Let’s discuss two topics.
First, why has the AI programming sector been so hot this year? Almost all internet companies, both domestic and international, are developing their own AI programming assistants. Last week, the media reported that Cursor completed a $100 million funding round, with its valuation increasing by 6.5 times in just four months. Incredible.
Second, I have been playing with Kuaishou’s Keling recently, giving it prompts to generate videos or turn images into videos, which is quite interesting. A colleague asked, since Kuaishou doesn’t have a cloud computing business and didn’t seem to invest heavily in large models last year, how are they now venturing into AI video services?
These are all interesting questions. After nearly a year of contemplation and discussions with many friends, I have only recently started to think clearly about them. Let’s start with programming assistants. Back in August this year, I spoke with the head of Wenxin Kuaima, Zang Zhi. At that time, I asked him what the difference is between having a large model write an article and writing code.
He said that code is a language of expressing logic, with very clear rules and boundaries. As long as the code adheres to standards and passes tests, it is good code, at least a 60% good code. However, writing is not like that; an article that appears splendid and well-structured can still be poor, and there is no way to test whether an article is good.
Code is like playing Go; winning is winning, losing is losing—it’s very clear. Writing is not.
Regarding what constitutes good writing, I like the viewpoint of writer Liang Heng, who says that a good article has three layers of beauty. The first layer is the beauty of description and narration, writing about scenery, objects, and events. The second layer is the beauty of artistic conception, conveying feelings, emotions, and sentiments. The third layer is the beauty of philosophy, expressing new thoughts and values.
Clearly, AI is not suited to any of these three layers. Silicon-based life forms are not adept at generating feelings and viewpoints, at least not at this stage. Of course, saying this does not deny AI creation; in fact, humans can completely optimize and iterate on the initial drafts created by AI.
AI is not good at generating viewpoints. This is my understanding of the boundaries of AI this year.
Many people understand the value of AI programming tools from the perspective of cost reduction and efficiency improvement. This is certainly one perspective. However, the trend I have observed recently is that products like Cursor are lowering the barriers to software development. A few years ago, I taught a beginner’s Python course. Back then, a user wanting to start coding had to learn the syntax first and understand the general principles of program operation before they could start coding, which could take at least a month for users without a foundation.
But now, with Cursor, the difficulty of getting started has been significantly reduced. I even feel that the emergence of Cursor has changed the learning path for programming. Next year, I believe products like Cursor will further advance in the Agent direction, and by then, the barrier for developing small applications (note: small applications) will continue to decrease. If this trend continues, can we consider Cursor as a low-code product rather than just a pure IDE?
This direction has a lot of imaginative potential.
Next, let’s discuss why Kuaishou would create a product like Keling. To clarify, my explanation may likely be a case of reasoning from results to causes, but you might check if the logic is correct. Historically, whenever the production barrier has been exponentially lowered, new opportunities have emerged.
One important factor that allowed short video products like Douyin and Kuaishou to stand out is the proliferation of mobile phones and the enhancement of mobile photography/videography capabilities. With a smartphone, you can shoot anytime and edit anywhere, which is much simpler than lugging around a heavy camera, editing footage on a computer with professional editing software, and then uploading it to Youku.
Now, with the maturity of AI, the barriers to video production may further decrease. A couple of days ago, I saw American director Jason Zada release a short film generated using Google Veo 2. He mentioned on social media that the entire creative process was completed using AI tools. I don’t know if it’s true or just hype, but I wonder, by 2025, AI-generated video technology will definitely continue to develop, and video quality will surely improve. Will a new world emerge then?
As long as the barriers and costs are lowered, new opportunities will definitely arise. This is a principle I firmly believe in.
Additionally, Li Kaifu recently discussed a not-so-new perspective: the PC era brought a computer to every desk; the mobile internet era made computing devices portable and allowed them to know who we are and where we are; the AI era empowers applications with smarter-than-human artificial intelligence, allowing each application to be rewritten. This change will explode this year and next.
AI is indeed a very important variable. I am also learning, thinking, and practicing. Yesterday afternoon, I launched a small column called “AI Trend Reference 2025.” I usually work on products and deeply observe AI, while MacTalk itself has a self-media attribute, so I often have opportunities to communicate with various people in the industry. This time, I want to write down my thoughts, the viewpoints I have observed, and the insights I gained from discussions with others, which may provide you with a little inspiration.
The current early bird price for this column is 99 yuan, and it will increase to 129 yuan after the early bird period ends. I update it quarterly, with one article each week. Let’s explore together what the latest product trends are domestically and internationally, what the latest viewpoints are, and what interesting products have emerged that the news media has not extensively reported on.
For example, I am currently writing an article on Agents. Over the past month, I have seen many similar discussions, and I want to present the meaningful viewpoints I have filtered through to you.
