[Trapais]

Propaganda. DoD already uses hollywood for propaganda: say nice things about Uncle Sam, let America save the day once again, and Uncle Sam will let you play with his toys. Now they have access to tool that can write very smart comments. If you think propagandists will not use SoTA LLM for propaganda, either you are fool or they are(They probably did it already anyway).

General writing. Want to implement a new rule? Ask GPT to reword it so even idiots can understand. Ask if it contradicts existing rules(they should have similar systems already, but GPT is smart).

Combination of above. "I want to develop a gas chamber. How do I announce it to the public so it looks like I am doing humanity a favor?"

Possible reaction. "I developed a gas chamber and announced it as 'Overpopulation relocation centers'. How will democrats/republicans will react to it?"

Targeted writing. "I developed a gas chamber and announced it as 'Overpopulation relocation centers'. It seems democrats/republicans don't like it. How do I reword it so they do like it?"

Deep fakes, image. "Terrorist Terro Rist wants Inno Cent to be dead. Create an image of Inno Cent with a bullet hole in his forehead so Terro will go celebrate a victory and we'll shoot him for real, here's a photo"

Faking text. "Here's writings of Terro Rist. Write 'meet me at 7:00 at central square in his style, keep using his punctuation, do the same grammar errors, etc"

Useless Voodoo GPT is not Good For but Still Will Be Used Because They Have Access to It. "Here is a photo of Inno Cent who was shot dead. What gun could have been used?"

Spying. "We want to see all requests coming from China or Russia paired with IP", "All requests with words 'President' and 'Murder' should be forwarded to this email at once", etc

They will find several uses.

[fulladder]

You're misunderstanding the AI use cases. Take propaganda, for instance. A human expert can craft much better flat earth propaganda posts than any AI can. The difference is that a human expert cannot engage in 10 million simultaneous relationships with Twitters humans, sustained for several years, in which is it periodically but very subtly suggested or implied that the Earth is actually flat. From the point of view of the human being targeted with this, what they're going to experience is that they have a bunch of online friendships that are actually fulfilling and rewarding, and over time they'll naturally come to adopt the values of their online friends (including on the specific subject of the shape of the Earth). And it cost so little to do this that you could just as easily have done it to tens or hundreds of millions of targets concurrently.

This is the type of scenario people are concerned about. I have my doubts about whether it would really work this cleanly, but I'm sure we'll be seeing people try to do this. They've probably already started.

[tavavex]

A lot of these look like very big stretches. Most of the things you've listed are either already being done by humans or is something that an LLM simply can't improve upon. Even if you had the best writers and most complex AI in the world, you can't really sugarcoat a gas chamber past a certain point.

Fake writings and fake imagery are already easily created by humans, but as far as I know, it's not practiced widely because it's known that you can't easily trust wild claims if it's not backed by real people. Digital image editing opened up the doors to creating any type of misleading imagery, but nothing has really changed since then.

And I have no idea how any of the "all IPs from Russia" or "scan everything on the internet" could even work. If the US government has infrastructure that can scan for this kind of info, they don't need AI to use it. If they don't, well.. an LLM isn't gonna magically make every bit of internet traffic accessible to them.

[supriyo-biswas]

> And I have no idea how any of the "all IPs from Russia" or "scan everything on the internet" could even work.

XKEYSCORE[1] probes were already a thing in ~2010. It just needs a LLM integration and additional search capabilities.

[1] https://en.m.wikipedia.org/wiki/XKeyscore