[eukara]

That's like comparing horse and carriage to a modern day truck though, no? This can be fully automated, generating photoreal content on its own. You couldn't airbrush a photo via a cron job like how you can now automate some model generating thousands of images of people rioting/looting for authoritarian purposes. Who will go through the effort of verifying every one of them? Other language models with precision issues?

[ModernMech]

> You couldn't airbrush a photo via a cron job like how you can now automate some model generating thousands of images of people rioting/looting for authoritarian purposes.

I get how this could be a problem, but it seems to me that it would only be marginally effective instead of exponentially as some assume. The reason I think this is because we already have this kind of thing going on without AI [1]. And while it does work, it's not clear to me if making the fake more realistic actually does anything for the kind of people who get worked up about this. If you want to claim Seattle is burning and it's not, just grab a picture of another place burning, and you're good. Narrative achieved.

If you can pass as authentic a photo from a different place taken at a different time, then what does it matter if you can generate a new one? Are you going to trick more people? Maybe. But I have a feeling the people who are most likely to get tricked, would have been tricked by far more mundane fraud.

[1] https://www.rollingstone.com/politics/politics-news/fox-news...

[hasanas]

People aren’t divided into “most likely to be tricked” and “least likely to be tricked”.

When you generate “photos” of Big Ben burning and throw it on Twitter, you will grab the attention of many more Londoners than you could with a random image of a burning house or whatever, many of which would at least take the time to verify. Just look at the Balenciaga Pope images.

By grabbing the attention of those you are essentially stepping into the realms of “historical alternations”.