Comment by tivert | Hacker News Reader

Comment by tivert | Hacker News Reader

Parent Full thread

> No, that's never happened before. You're crazy.

Start filling up Discords with insane AI-generated garbage, and maybe you can devalue the data to the point it won't get sold.

It's probably totally practical too, just create channels filled with insane bots talking to each other, and cultivate the local knowledge that real people just don't go there. Maybe even allow the insane bots on the main channels, and cultivate the understanding that everyone needs to just block them.

It would be important to avoid any kind of widespread conventions about how to do this, since and important goal to to make it practically impossible to algorithmically filter-out the AI generated dogshit when training a model. So don't suffix all the bots with "-bot", everyone just need to be told something like "we block John2993, 3944XNU, SunshineGirl around here."

If we work together, maybe we can turn AI (or at least LLMs) into the next blockchain.

The equivalent of "locals don't go to this area after dark"? I instinctively like it, but only because I flatter myself that I would be a local. I can't see it working to any scale.

> The equivalent of "locals don't go to this area after dark"? I instinctively like it, but only because I flatter myself that I would be a local. I can't see it working to any scale.

I was think it could work if 1) the noise is just obvious enough that a human would get frustrated and block without wasting much time and/or 2) the practice is common enough that everyone except total newbies will learn generally what's up.

> The equivalent of "locals don't go to this area after dark"?

We have this with human online places already it's called 4chan

This idea has been talked about enough that we call it "Habsburg AI". OpenAI is already aware of it and it's the reason why they stopped web scraping in 2021.

This is an intellectually fascinating thought experiment.

> This is an intellectually fascinating thought experiment.

It's not a thought experiment. I'd actually like to do it (and others to do it). IRL.

I probably would start with an open source model that's especially prone to hallucinate, try to trigger hallucinations, then maybe feed back the hallucinations by retraining. Might make the most sense to target long-tail topics, because that would give the impression of unreliability while being harder to specifically counter at the topic level (e.g. the large apparent effort to make ChatGPT say only the right things about election results and many other sensitive topics).

Reply on news.ycombinator.com