3 Comments
User's avatar
тна Return to thread
Alberto Romero's avatar

It can't without living like a human, being like a human. That's what I'm saying.

Expand full comment
Phil Tanny's avatar

I write an article about my experience of Covid. What stops some version of AI from successfully imitating that article, with "success" defined as most readers not realizing it's AI generated?

Expand full comment
Alberto Romero's avatar

I don't define success like that - that's deception. It might be easy to make people think an article has been written by you when it's AI-generated instead (although I'd say that doesn't say anything good about your writing!) What I'm saying is that it's still easier to feel like a person on the other side of the screen in general, not just in the articles you write, also in these kinds of exchanges.

But even if I accepted your premises I'd say: Try. Try to make an article with ChatGPT feel like one written by a person. You will have a hard time because these systems, the ones that exist today, tend to linger around the center of the distributions of what's possible to say. That's why the writing feels bland, dull, and full of generalities. Humans, in contrast, write about what's unlikely, and implausible. We behave very differently.

That's the issue with your thought experiment, that you make the jump from "I wrote X article" to "what stops an AI from writing X article" but in practice, you find that there are actual things that stop it from doing so - not because no one will be fooled but because you can access that place I refer to and AI can't.

Expand full comment