Einige Claude-Benutzer sind sauer, dass die neuen Wasserzeichen von Anthropic sie dazu bringen, es bei ihrer Arbeit oder im Unterricht zu verwenden
Einige Claude-Benutzer sind sauer, dass die neuen Wasserzeichen von Anthropic sie dazu bringen, es bei ihrer Arbeit oder im Unterricht zu verwenden
25 Kommentare
Losers not users… The problem with reliance on AI to do your writing, is that it makes strong minds weaker, and weaker minds “appear” stronger. Neither of which is good for the species as a whole. This should be law, not voluntary by one org.
they will just paste it into another ai and ask it to reword it, quality is dropped and efficiency is lost.
The smooth brain cohort is upset.
A whole article written about a handful of mostly downvoted reddit posts and comments.
Really scraping the bottom of the social media barrel today, huh?
The bigger problem people are not addressing is whether people can actually verify AI output.
If we are going to use AI, humans still need enough expertise to know when it is wrong, question what it gives us, and verify the information.
We cannot just assume an answer is trustworthy because it sounds confident. That means we still need experts, or people who know when to bring experts in, for AI output.
My job encourages use of AI in practically all aspects of our work. So the watermark is just redundant. Moot issue for me.
Will be intetesting to see how many job postings are made with AI but annoying if I can’t apply with it in turn.
Now why would they be mad about that? Isn’t authenticity important?
There’s already open source packages aimed at neutralizing any such watermarking.
Lucas Ropek got really personal in this article – blasting a single Reddit user to make his point. Cherry picking comments to promote a narrative isn’t very honest journalistic work, Lucas.
Won’t people just start running open weight models on their macbooks?
What if your job blindly accepts the slop dripping from users keyboards? The place I work they encourage it and everyone sends direct messages to each other with AI.
Do we know what this alleged watermark looks like in code?
It’s so obvious when people use AI at my work. If I see bold underlined words and circular reasoning I know it was thrown through AI. If I see celebratory remarks while stating negative performances I know it was done by AI. Makes the worker (usually people above me) seem robot as well.
Maybe don’t rely on AI to do your work for you.
Notice how it’s only the bad workers who are against this.
Really? That’s the article? It is a nothing burger rehash of some reddit threads, it is not even touching on the discussions of the mechanisms of the likely Synth-ID inspired system, or more insightful (if broadly critical) takes from LocalLLaMA.
For my own sake I find the concept of watermarking in that manner creepy, and wonder about the particulars of Anthropic’s implementation (as they’ve not given full details) on low entropy tokens situations like math, coding, or codons, where accuracy may be paramount. That said, use abliterated local models and host your own tools!
I’ve been using AI to fight Flock cameras. All kinda FOIA requests, constituent outreach letters, and feedback on open legislation.
Luckily the government will probably still be too dumb to figure out that its mostly AI written.
I like how the whole article is about comments on a Reddit post, which is now reposted on Reddit to comment on…
“Catch them”?
My job MANDATES we use it.
This won’t impact me at all
Would it not be possible to remove these watermarks either through another program that can see them that I’m sure will quickly pop up or through just pasting it in notepad, etc?
My job wants me to use AI as much as possible. “Getting caught” might get me a promotion.
They don’t need to worry we can already tell
One of the biggest problems with AI is that nobody double-checks it. It wouldn’t be so problematic if someone actually checked its output. So good on a watermark because it will actually force people to check the content.
Oh, also good on the watermark because AI-generated content isn’t copyright protected. Anyone passing off AI content as their own work deserve to have other people take it and use it free of charge.
An article about reddit comments, wild.
Th best watermark for Ai content is simply asking the person presenting it about it.
Youll see pretty quickly they habe no clue.
Its an illusion and people hate the watermarks are z reminded they couldn’t make this without claude. That is why certain people are so pissed