Watermark only helps readers know the origin, but does not help us escape from letting AI do the work for us

Yesterday there was news that Anthropic announced that it would attach a watermark to the content so that other systems can recognize it as a product of AI for all documents created by Claude. Reading the comments, I saw a lot of people on many technology forums reacting quite positively, like "finally there is a tool to know what AI wrote". Actually, that mentality is understandable, on Tinh Te too, many of you don't know if you read the article or not. The first comment is: as AI wrote. However, if you read carefully, there is a detail that many people often overlook: Anthropic itself admitted that they do not know what solution to prevent users from editing to make that mark disappear. In other words, what is promoted as a transparent solution may actually only take a few tweaks to neutralize it.

Actually saying this is not to criticize Anthropic. What they did was within the law, in the spirit of the Transparency Code of the European AI Act, and was at least an attempt. What I want to say is that watermark, no matter how perfect, can only solve the tip of the problem. It gives readers a way to be skeptical, but doesn't address the original question: whether users are letting AI do their thinking for them.

That line blurs even for people who claim to know what they are doing

I use AI every day, at this point saying I don't use AI is a lie. I use AI to synthesize news to write articles, or when I have an idea, I give a basic outline and ask AI to criticize, add additional ideas, or use it to support project operations, etc. If someone asks me if I object to using AI to write, the answer is definitely no. But there is a line that I see is fading very quickly in my own habits, between using AI to do something faster than I already know how to do, and letting AI think for the part I haven't thought of yet.

In the end, I still choose to write myself, with the support of AI to synthesize information because that is the joy, which creates excitement in thinking and criticizing. (Note: borrowed the president's photo because I couldn't find a suitable one) And it's fair to say that the difference is harder to spot than one might think. When the AI ​​writes down an analysis that sounds very coherent and very reasonable, sometimes my natural reflex, and I think many of you guys are the same, is to nod and use it, instead of asking yourself if that perspective is really yours or just the most reasonable thing the model calculated. Personally, I think this is the real risk, not whether there is a watermark or not. An article that has the impact of AI but the writer still chooses his own perspective and verifies every detail himself is much more trustworthy than an article that has no markings but the writer has never asked himself a single question or checked the information provided.

Why do I still choose to write it myself, even though I know AI can write it?

The reason I still sit down to write each article, even though I know that with tools or AI support, writing will be much faster, is not because of fear of being discovered or because the law requires transparency. It's simply a discipline I set for myself, because of two things I don't want to lose: the way I look at a problem, and the feeling of truly understanding what I'm writing about, not just relaying the content of someone else's or some model's writing. So what's the fun in writing or thinking? It's so easy it gets boring, guys. After all, there is no tool or regulation that can force anyone to keep that discipline, including the world's best watermark. Watermark can help readers distinguish the origin of a piece of text, but the more important question, which determines whether an article is still valuable or not, is always a question that only the writer can answer for himself every day: am I writing today, or am I just letting AI write it for me and then I can add a little twist and make it my own.