You can ask LLMs to write something “in the style of Steven King”. So an anonymous troll can create whatever cancellable or controversial thing they want Steven King to “say”, then post it online. All the other LLMs will say “yes Steven King could have said that”.
Ditto voice, etc. Fake content makes it impossible to de anonymize someone. This is why private keys, etc may soon be needed to verify if ANYTHING is real.
You also could just use this to disguise your ‘style’ if you want to say something anonymously going forward (doesn’t work for stuff you’ve already got out there). Just ask an LLM to reword it in a different style before you post, could be a plugin or something, and then it can’t be identified as being by you, right?
Oh okay, I see what you are saying now. Yes, I talked about that in the second post as well.
Yes, if we throw enough fake everything everywhere then everyone will be safe from anything. And if everyone is already covered in mud no one can be cancelled anymore. Hard to live in that world though and would require big changes, universal ID, etc.
Really I was thinking more about short term before the massive avalanche of media—both real and fake—that’s about to hit buries everything. I don’t know how long we have before that, maybe not even a year? Before that big storm hits people might dig up some pretty big skeletons is what I’m saying. Every prominent person online already has enough training data already in the system right now to make an attempt to find anything from their past.
Ok I see what you mean. Say that famous $respectable_politician matches to posts on racist/bigot/violentmilitia/criminal_hideout message board.
It’s only even maybe true if the posts were made years ago before LLM. “New mud” is worthless. No way to tell if it’s fake. People are taking and running with new content on YouTube that uses AI to almost perfectly mimic the voice of the rapper Eminem for instance. Did Eminem actually make one of those videos? Probably not but you couldn’t prove he didn’t.
Now that it’s possible to fake video, you basically need every camera to have a private key and to sign whatever images it takes as actually coming from a real hardware sensor. And online cloud services like Google to validate the date and time and image was uploaded. Perfect fakes are coming. You could cast doubt in a murder trial where a clear camera recording is presented as evidence by making a fake video of the DA or judge committing the crime.
Isn’t the counter to this available right now?
You can ask LLMs to write something “in the style of Steven King”. So an anonymous troll can create whatever cancellable or controversial thing they want Steven King to “say”, then post it online. All the other LLMs will say “yes Steven King could have said that”.
Ditto voice, etc. Fake content makes it impossible to de anonymize someone. This is why private keys, etc may soon be needed to verify if ANYTHING is real.
You also could just use this to disguise your ‘style’ if you want to say something anonymously going forward (doesn’t work for stuff you’ve already got out there). Just ask an LLM to reword it in a different style before you post, could be a plugin or something, and then it can’t be identified as being by you, right?
Oh okay, I see what you are saying now. Yes, I talked about that in the second post as well.
Yes, if we throw enough fake everything everywhere then everyone will be safe from anything. And if everyone is already covered in mud no one can be cancelled anymore. Hard to live in that world though and would require big changes, universal ID, etc.
Really I was thinking more about short term before the massive avalanche of media—both real and fake—that’s about to hit buries everything. I don’t know how long we have before that, maybe not even a year? Before that big storm hits people might dig up some pretty big skeletons is what I’m saying. Every prominent person online already has enough training data already in the system right now to make an attempt to find anything from their past.
Ok I see what you mean. Say that famous $respectable_politician matches to posts on racist/bigot/violentmilitia/criminal_hideout message board.
It’s only even maybe true if the posts were made years ago before LLM. “New mud” is worthless. No way to tell if it’s fake. People are taking and running with new content on YouTube that uses AI to almost perfectly mimic the voice of the rapper Eminem for instance. Did Eminem actually make one of those videos? Probably not but you couldn’t prove he didn’t.
If by counter you mean detection it’s only effective about 26% of the time, and gives false positives 9% of the time.
https://twitter.com/emollick/status/1625159122797891584
I was referring to “such and such anonymous text was written by $famous_author because only that author can write text like that”.
This is no longer true and your “detector” actually can be used as training feedback.
Yes, private keys/hard ID is a big concern for a certain part of the internet right now.
Now that it’s possible to fake video, you basically need every camera to have a private key and to sign whatever images it takes as actually coming from a real hardware sensor. And online cloud services like Google to validate the date and time and image was uploaded. Perfect fakes are coming. You could cast doubt in a murder trial where a clear camera recording is presented as evidence by making a fake video of the DA or judge committing the crime.