[lolinder]

Yeah, there's a lot of confusion between AI training and AI agent access, and it's dangerous.

Training embeds the data into the model and has copyright implications that aren't yet fully resolved. But an AI agent using a website to do something for a user is not substantially different than any other application doing the same. Why does it matter to you, the company, if I use a local LLaMA to process your website vs an algorithm I wrote by hand? And if there is no difference, are we really comfortable saying that website owners get a say in what kinds of algorithms a user can run to preprocess their content?

[jsheard]

> But an AI agent using a website to do something for a user is not substantially different than any other application doing the same.

If the website is ad-supported then it is substantially different - one produces ad impressions and the other doesn't. Adblocking isn't unique to AI agents of course but I can see why site owners wouldn't want to normalize a new means of accessing their content which will inherently never give them any revenue in return.

[lolinder]

I don't believe that companies have the right to say that my user agent must run their ads. They can politely request that it does and I can tell my agent whether to show them or not.

[jsheard]

True, but by the same measure your user agent can politely request a webpage and the server has the right to say 403 Forbidden. Nobody is required to play by the other parties rules here.

[lolinder]

Exactly. The trouble is that companies want the benefits of being on the open web without the trade-offs. They're more than welcome to turn me down entirely, but they don't do that because that would have undesirable knock-on effects. So instead they try to make it sound like I have a moral obligation to render their ads.