|
I can recognize it too. For example, there's that particular archetype where the writing summarizes things by leading with a block quote. But I don't think that's a bad thing. To be honest, I actually think some AI style of speaking is effective for conveying information. People can have different opinions, of course. I write most things with my own hands. But if I feel like the AI's suggestion is better than what I wrote, I'll improve my writing. In fact, I did exactly that with a few pieces recently. I respect that some people prefer things unpolished, and I respect their way of thinking. But I don't think that flat, AI assisted writing is bad for information delivery. I believe there are many different kinds of writing. In novels or reportage, I think AI writing is bad. The style itself is part of the author's signature, after all. But on the flip side, I'm skeptical about whether someone's personality really needs to come through in a wiki or an encyclopedia. The core of the anti AI argument right now is ultimately about form and style. But that style is something the model was trained on, based on datasets that people, on average, preferred. In other words, it's the most average, the most flat. It's inorganic, but it's the median. When we define readers or consumers, there's always the question of where to draw the line, what segment to set. And when you think about it, the AI's training data comes from the broadest consumer distribution. It spits out that kind of data because it learned that most people preferred it. Of course, even in informational writing, things like "has this person actually been through this?" or "is this exaggerated?" do matter, and that feeling is important. But when I read a technical article, the most important thing, unlike a novel, isn't the prose style. It's whether it hands me the technique or not. So while I respect your opinion, I don't think it aligns with my values. |
That's not what it is! The heavy reinforcement learning that the models go through makes their writing the farthest thing possible from the "most average, flat" distribution. The distribution actually becomes very sharp due to this, unlike any writing it has seen in 99.99% of its training data! (The heavier the RL is, the more same-y the output becomes.) This is why every AI-written piece feels the same, and why people can learn to easily tell that something was written by an AI. And frontier labs make this problem even worse by how they sample[1] their logits.
[1]: https://gist.github.com/Hellisotherpeople/71ba712f9f899adcb0...
> But when I read a technical article, the most important thing, unlike a novel, isn't the prose style.
Exactly! Which is why I think there's no need to "enhance" it with AI.