Unless we can find some way to verify humanity for every message.
A mix of social interaction and cryptographic guarantees will be our saving grace (although I'm less bothered from AI generated content than most).
There is no possible way to do this that won't quickly be abused by people/groups who don't care. All efforts like this will do is destroy privacy and freedom on the Internet for normal people.
So we need some mechanism to verify the content is from a human. If no privacy preserving technical solution can be found, then expect the non-privacy preserving to be the only model.
There is no technical solution, privacy preserving or otherwise, that can stave off this purported threat.
Out of curiosity, what is the timeline here? LLMs have been a thing for a while now, and I've been reading about how they're going to bring about the death of the Internet since day 1.
It’s slowly, but inexorably increasing. The constraints are the normal constraints of a new technology; money, time, quality. Particularly money.
Still, token generation keeps going down in cost, making it possible to produce more and more content. Quality, and the ability to obfuscate origins, seems to be on a continual improve also. Anecdotally, I’m seeing a steady increase in the number of HN front page articles that turn out to be AI written.
I don’t know how far away the “botnet of spam AI content” is from becoming reality; however it would appear that the success of AI is tightly coupled with that eventuality.
I give it a decade. By that time social media had done irreparable damage to society.
And now people are receiving generated emails. And it’s only getting worse.
