The Videogamer thing is a good illustration of why AI specifically fails at criticism. Next-token prediction and RLHF optimize for typicality, so you end up with text that reads like an average of all reviews ever written. You can't get sharp, weird, idiosyncratic takes out of that process. I wrote about that convergence-to-the-mean problem here: https://philippdubach.com/posts/the-impossible-backhand/
But I think the harder part is that enforcement depends on detection, and detection will always lag generation..
This seems totally reasonable. Even if we say AI can play games and watch movies and have opinions about them, which I don’t think they can, their opinions aren’t super relevant to humans.
It's one outright AI ban that I actually agree with.
I can see some beneficial purpose for almost all forms of AI, but allowing AI generated reviews (for anything) is a perversion of the concept of a review for humans. Besides, you know 1000% AI reviews are going to be gamed and steered by humans running the operation for profit, so it's not really an AI review anyway, it's just ai-washed human fraud.
The day I will accept AI reviews is the day we declare AIs to have human rights and responsibilities. Until then it's just slop and/or lies, depending on how charitable we're being.
The Videogamer thing is a good illustration of why AI specifically fails at criticism. Next-token prediction and RLHF optimize for typicality, so you end up with text that reads like an average of all reviews ever written. You can't get sharp, weird, idiosyncratic takes out of that process. I wrote about that convergence-to-the-mean problem here: https://philippdubach.com/posts/the-impossible-backhand/
But I think the harder part is that enforcement depends on detection, and detection will always lag generation..
This seems totally reasonable. Even if we say AI can play games and watch movies and have opinions about them, which I don’t think they can, their opinions aren’t super relevant to humans.
It's one outright AI ban that I actually agree with.
I can see some beneficial purpose for almost all forms of AI, but allowing AI generated reviews (for anything) is a perversion of the concept of a review for humans. Besides, you know 1000% AI reviews are going to be gamed and steered by humans running the operation for profit, so it's not really an AI review anyway, it's just ai-washed human fraud.
The day I will accept AI reviews is the day we declare AIs to have human rights and responsibilities. Until then it's just slop and/or lies, depending on how charitable we're being.