the claude watermarks don't even work that well, but they are 100% worth it in pissbaby piss
https://bsky.app/profile/thelincoln.bsky.social/post/3mzba4cdtuc2z
1 Boost
0 Likes
@davidgerard Interesting. I'm curious how this is actually supposed to work. I would guess that it only applies to those who copy-paste from Claude but not people who just retype what Claude spits out.
Sadly, my tiny violin is now so small that i can't even find it with an electron microscope
@davidgerard@circumstances.runYou understand their point tho - they're paying for a plagiarism machine to tell them what to think and it's completely unfair to expose that they don't think for themselves anymore. Just ludicrous that they can't get paid for a chain of words produced by a statistical language model
![[object Object]](https://media.mas.to/accounts/avatars/109/287/758/133/964/366/original/0f3099c0adfb36a0.png "[object Object]")
@davidgerard the “how can you ban LLM code when people can just lie about not using LLMs” guys when there’s the thinnest chance they won’t be able to just lie:
Also, from my experience, these fuckers *really* don't need a watermark anyway. They *never* hide the fact they love their toy and use their toy. If they do try and hide it, the hiding won't last long before they come out swinging with "you know... I am pretty critical of LLMS... but..." @zzt @davidgerard
@WeirdWriter @zzt how can you tell if someone used the chatbot? they'll tell you, every time
How do watermarks work, especially on plain text where one could -- in principle -- rewrite it completely?
@weekend_editor @zzt @davidgerard i'm not sure, but it's my understanding that there will be "watermarks" in all levels of the generated content, from the content and structure of the text down to the choice of words and punctuation
in any case, people using ai to generate text are using it precisely because they don't want to bother writing the whole thing in the first place, so why should we believe they'll rewrite it once it has been generated?
You guys are asking a sociological question of whether slop-users will actually rewrite or even rework what they get back. That's an important question, and it's definitely one we should all be asking.
Just... don't ask it of me, because I have approximately the social skills of a potato.
I was asking a *technical* question: how in the world can one watermark a text file, especially if it's straight ASCII?
1/n
@davidgerard
I thought the AI positive crowd was proud that they used AI for everything?
@davidgerard Interesting. I like that it's not something stupid where it only applies to copy-pasted output or something like that.
Also completely unsurprising that they could've had this implemented all along but are only doing it now because the EU has said they have to.
https://www.anthropic.com/news/claude-text-watermark
How Claude's text watermarking works