Comment by bigyabai
15 hours ago
I can't fully understand this mindset. If you despise AI watermarks, why not rent your own inference hardware and use open-weight models without watermarks? Or LORAs that remove them?
AI steganography exists because all customers are a liability. These AI companies don't really care if you get uncensored or use un-watermarked models, they just want to limit their personal liability when you use their service. If a criminal's chatbot logs get entered into a case as evidence, being able to exonerate your service could save hundreds of millions in legal fees. It's like Apple/Google/Microsoft's CSAM scanner on their cloud services, it's a measure that exists for the benefit of the court and the company.
The problem isn't the government, or the corporate relationship with the justice system; you can't change either of those without overriding Rule of Law. The problem is exactly what people said upthread; centralization. If you refuse to use anything but the most mainstream smartphone, mainstream social media and mainstream AI provider, you're probably opting in to a panopticon no different from Xkeyscore 20-odd years ago.
If you really hate AI surveillance, just host it all yourself. OpenAI and Anthropic were never going to be your savior, it's exhausting going through the motions of "oh no, they're a federal honeypot in disguise!" every time basic retention infrastructure is discovered. The writing has been on the wall for so long that I can't empathize with people that think this is the straw that will break the camel's back.
> If you really hate AI surveillance, just host it all yourself.
If you think this is a solution for 99.999% of the world, I've got a Linux on Desktop fantasy to sell you.
The world will never have distributed Facebook or Matrix or whatever pipe dream nerds think will work.
The only way to defend ourselves is to protest and legislate against this madness. And unfortunately the political willpower is on the side of surveillance.
Local AI will never be a mainstream thing.
Well yeah, those first two lines have been true for the past 25 years and AI doesn't really change the shape of the problem.
> The only way to defend ourselves is to protest and legislate against this
You can protest and legislate against it, only to have the ruling reversed in a sealed court decision or reactionary "Patriot Act"-style legislation. It doesn't really defend anyone.
The only surefire solution is to stop using remote services. You will never be able to control what a server serves you, so there is an inherent mistrust in demanding certain features be added or removed. When people make these sorts of high-and-mighty demands, that's when we see security theater, theoretical whitepapers and dishonest threat modelling from these businesses. For all of the effort they go through to ensure that you can trust them, you'll never get a copy of the program they run on their server. And that's by design.