@tjrobertson52

I keep coming across this idea that pretty soon all content is gonna be AI generated. People won't have a reason to make original content, and therefore, these AI models are cannibalizing themselves, because, of course, they need original content to continue training. And if we remove the incentives to create that original content, they won't have it to train on.

And I actually think this is nonsense. So I just want to push back against this idea. First I'll take the easy version, the straw man of the argument, and then I'll do my best to steal man the argument, and still, I think, take it down.

So the easy version is taking their claim at face value, no one will be creating content. So let's just imagine that world we're in the future. AI contents everywhere, and literally everyone stops creating original content.

These companies desperately need original content to train on, and there's nothing to be found. I think it's pretty easy to see that in this world, there would actually be some huge incentives for creating original content. Not only would the AI.

Companies be willing to pay for it, but if you are one of the only people creating original content, you would have a ton of influence over how these models respond. Now, of course, an intelligent person making this argument would probably respond, okay, sure. I didn't mean no original content.

There will just be less original content, and that'll be bad. For the world and bad for these AI. Companies.

But I don't think that's true either. I think it's important to consider why these AI companies need original content. Do they really just need more words that haven't been strung together in a particular order?

No, they could easily generate that themselves. Do they need more examples of good sentence structure? No.

They have trillions of pages of that. The actual value in original content is in new information. So the question is, will people be sharing new information less frequently in this world dominated by AI.

Content? All the evidence I can see points at new information sharing accelerating. It is already so much easier for someone with truly novel, insightful ideas to generate and distribute content representing those ideas.

And yes, I do mean generating content with AI. Then you might say, aha! See, it's just gonna be more AI.

Slop. The AI. Can't just train on AI.

Slop. But when this is done well, the only thing AI. Is contributing to is the actual sentence structure, which, again, that's not the thing the AI.

Needs when we're talking about original content. Today, I create hundreds of times as much original content online as I did three years ago. It's almost all AI generated, and I've never had someone say, this looks AI generated.

Over time, more and more people are gonna learn how to do this, and it's gonna Get easier and easier to do, and anyone with original ideas or information is going to be able to create content at scale that's incredibly valuable to these AI models. So while I do think real harm is going to be done because of this, I think it is inevitable. And this idea that original content is going to dry up is just wishful thinking by people who wish it weren't so.