There are several ways to produce a usable news poster like the one shown above:
Purely Manual: Of course, this description isn't entirely accurate, but let's call editing in Word-like document tools "purely manual." Every character is typed via keyboard, and the layout is adjusted manually.
Semi-automated: If this is not a one-off task but a daily routine, it can likely be standardized into a series of templates and tools to improve efficiency.
Fully Automated: Automatic news crawling, automatic application of templates and tools, and automatic output—all driven by programs in the background.
AI (Agent) based: Generated by a model. Yes, the image above was generated through the updated GPT-4o with just two simple rounds of conversation.
In fact, in terms of efficiency alone, there might not be a huge difference between "3" and "4," but the underlying logic is completely different.
If the target webpage structure changes or the data template requirements change, it might take a significant amount of time to modify the code. Furthermore, if the programmer who originally developed the "fully automated" system has left, the person taking over might spend even more time or discover more bugs.
We need AI (Agent) models because we need the model to complete the task autonomously even when the aforementioned changes occur.
Admittedly, at this point in time, the poster generated above still contains quite a few errors. Compared to human-produced content, there is still significant room for improvement in quality. However, if we look at the exaggerated pace of progress in model capabilities over the past two years, there is every reason to believe it will get better at an accelerating rate.
I used to think that everyone would want "4," but now I am increasingly finding that the choices are far from being so one-sided.
For me, for quite some time now, I have hoped to delegate as much work as possible to models. Only through constant experimentation can we understand the boundaries of capability at any given time and predict the possible evolutionary directions of these models.
Therefore, in just over an hour this morning, starting with a few keywords, I used deepresearch + cursor to rebuild a knowledge graph and interactive website for the AI industry chain as perceived by the model.
I know that humans could do it better, but I also know that in the not-too-distant future, humans will be surpassed in this task as well.
Because, from form to content, everything is getting better at an accelerating pace. This is a world increasingly driven by computation.