Hacker News Reader

Comment by kylebenzle | Hacker News Reader

[flagged]

the ease and scale of modern tech is on a completely different level to 2004 photoshop

And the scale and quality of 2004 photoshop was unlike anything before it.

And the rise of photoshop use had created a shift in how we perceive images as a proof of events. It had also changed people's concerns about the privacy of personal pictures exactly because of this concern. This tech takes it to the next level of ease, spread, and eficacy to the point of video (which were nearly impossible a decade ago). The fact that significant technological shifts happen doesn't mean they aren't any less impactful.

A hydrogen bomb is more powerful than a nuclear bomb is more powerful than TNT. That doesn't mean the hydrogen bomb isn't a step-change in the danger level.

Differences of quantity can become differences of type. It's one thing for instance of a handful of governments have nuclear weapons, it's another if a few million individual can get their hands on them.

I think it's not really clear if the AI flood is just like Photoshop, and what the long-term effects are.

AI: both a game changing revolutionary technology and "nothing."

I mean, it's probably markedly less fun to see your exact likeness participating in pornographic acts, but, the thing is, you're right on the money here because this is the very tip of the iceberg of the horrors that are forthcoming with AI.

I hate to say it, but Taylor Swift specifically hasn't seen anything yet.

Question is, how can it be blocked?

An image is quite unique in it's hash/signature, but a video generate by AI can have billions of unique seeds.

A computer can generate billions of unique seeds, but not billions of relationships with your sales team.

Maybe don't allow companies you don't trust to buy ads?

If only companies with a relationship with your sales team can buy ads from you, and they know they'll lose that ability if they run a fake AI ad, then you have far less to worry about.

It could be argued that this may work for newspapers, TV, and radio stations, but can't scale for serious websites. Serious websites won't be taken seriously if they can't get this under control, and folks will be turning back to newspapers, TV, and radio stations to get away from it. Don't make them competitive again.

For blocking you could use AI to determine if this is a picture of a celebrity that is nude. You'd probably whitelist the nude pictures of celebrities that have intentionally been photo'd nude. It's another question if any of the social media sites out there actually want to do this. Meta might, they seem interested in that type of research.

The only way this will effectively be dealt with is:

1. Banning the people caught doing it from social media.

2. The law enforcement/judicial way.

I don't know if new laws are required, there's probably something that already applies. E.g. if you Photoshop people in various ways and distribute the images, I'm quite sure the perpetrators get punished.

Bing already is very sensitive to using real people in its images though it's easily worked around. Ask it to draw Abraham Lincoln sinking in quicksand, it's likely to refuse. Ask it to draw a "presidential looking tall man with a dark beard but no moustache, dressed in a dark suit with a bow tie and a stovepipe hat" sinking in quicksand and it's going to make someone who looks like Lincoln far more than just that prompt should have lead to it creating. For someone like Taylor Swift, there are probably many attributes of hers that could be included in a prompt that would cause it to zero in on her likeness without it knowing that it is her that it is drawing.

I haven’t really played with Bing, but using some other site I was simply able to say a “a person who looks like _______” and it would get really close, while saying to use them directly wouldn’t work.

> For blocking you could use AI to determine if this is a picture of a celebrity that is nude

While it would probably work, it sounds prohibitively expensive.

They’ll probably just criminalize possession of GenAI image models. That’s clearly where the political winds are blowing.

This would be struck down in violation of the First Amendment. One has a right to make images of others in anyway they see fit. If they try to sell their likeness in a commercial transaction, then things can be regulated. But if someone is releasing artwork for free of someone else, that is fair game.

> But if someone is releasing artwork for free of someone else, that is fair game.

Within limits. If, say, said artist posts from their own account, signs the art, or otherwise links the deepfakes to themselves, then this could be unauthorized use of someone's image in order to benefit by raising their own status or otherwise profiting in another way than direct sales. I'm sure that's actionable.

You run the image through something like CLIP interrogator.

Write laws that only allow early movers like Microsoft/OpenAI to have legal access to AI technology. Anyone else who wants to use them will have to go through one of these early movers who are granted exclusive license by the government and decide which uses are permitted. That won't stop some from still doing things on their own but will allow the government to imprison those caught doing it without authorization.

Given how easy it is to run LLMs at home, it seems that maybe it's time to also roll back the failed PC experiment and require permits to run computers at home with strict usage type limits.

Why not just give these mega corporations complete control over the whole internet, while we're at it?

It's notable that Microsoft has already come out and said the tech industry needs to write laws for Congress to pass on this matter. They were waiting for this moment to happen and are ready to take advantage.

Reply on news.ycombinator.com