After One-Click Video, Ask Yourself One Question First
The English is translated from the author's Chinese. 中文版
Our team has made short science videos since 2013. In the past, one video needed a few people who were good at different things: planning and writing, art design, editing, and visual effects. They worked together for three to five days.
Over the years, our skills grew across these lines. Our writers learned to edit video. Our designers learned to read science papers and check facts. But in the end, a few people with different strengths, working together, still made videos fastest and best. So in the past, not many people could make videos, and not many videos were made. Few people ever needed to ask: "What kind of content am I making?"
Now things are different. After Opus 5.5 was released on September 22, the old rules of video making no longer hold. Using code to make videos has become very popular. A person who has never made a video can subscribe to a top model and make twenty videos in one day. Their quality is not clearly worse than mainstream science videos from around 2016.
Making videos is no longer a rare skill. Anyone can do it, and this cannot be stopped.
So a question that did not matter before now matters a lot: Am I offering work-type content, or service-type content?
Two kinds of content
People watch service-type content to meet a need: to understand something, to pass ten minutes, to learn how to use a thing. People watch work-type content to enjoy the content itself: a TV series, a novel, the special way one creator tells a story.
Here is a comparison. When the food industry grew, ready-made ingredients let people who cannot cook open breakfast stalls. A breakfast stall sells a service: food that fills you up, and is clean, fast and cheap. Nobody expects it to win on cooking skill. A breakfast stall and a restaurant are both honest businesses. They just win by different rules. If a stall owner runs his stall by restaurant standards, he will work hard where no customer pays for it, and get careless where the business is really decided.
These two terms come from our paper. What follows, about why it may help to know the difference, is our own extension.
Why it may help to know the difference
If you make content. First ask yourself which kind you make. Service-type content competes on whether the need is met, whether it is reliable, and how much it costs. How beautiful it looks matters less. And anyone can come in, so this road will only get more crowded. Work-type content competes on whether you have something to say. AI makes both kinds faster to produce, but it does not create more people who have something to say. With the same tools, a beginner will find it hard to beat someone with more than ten years of experience at choosing topics and making judgments. And people like us, with more than ten years behind us, also need to think clearly about which side our skills belong to.
If you build tools. Most users of one-click video tools make service-type content. So such tools should perhaps be judged by whether users' needs are met: Is the output reliable? Is it easy to edit? What does it cost? If you sell or judge these tools by their "creative ability", you may be aiming at the wrong target.
If you build models. Making a beautiful explainer video and writing a story that people want to keep following are two different abilities. If one benchmark gives them one score, the difference may disappear. If you test them separately, it may be easier to see what a model is really good at.
Finally
One-click video tools make exactly service-type content.
Our paper discusses a few things this short article does not cover: why the ratio between the two kinds slowly becomes unbalanced; why the recommendation pool on free platforms, which picks content through layers of trial viewing by users, stops working once content grows past a certain multiple; and one possible way out, which we call the BoLe system.
If you want to see how the recommendation pool fails, open our demo. Set AI output to 100x and watch one year go by.
We hope this paper is a little helpful to you. Corrections are welcome.
