на ху я?
How Claude marks AI-generated content
311–320 of 446 posts
Re: How Claude marks AI-generated content
#312Is the detection mechanism going to be open, free, and possible to run locally without prostrating to an opaque third-party company that will do whatever they want with the text content provided (including using it for training), and take no responsibility in case of false-positives for which there can exist no proof or evidence against by the victim? This is another useless, if not actively harmful, performative EU…
[dead]
Re: How Claude marks AI-generated content
#313It's inevitable there will be false positives, inevitable they'll do reputational or economic damage, and inevitable plaintiff attorneys will sue on the behalf of people damaged. Making it worse, it's product liability blended with defamation. Anthropic should've told the EU to pound sand and geofenced off Claude. If they don't want to live in the dark ages, elect smarter people.
Re: How Claude marks AI-generated content
#314Earlier quoted context omitted.
Your "code that Claude makes..."? Oh, how I laughed. That was never your code, my friend.
Please don't let the arbitrary selection of phrase distract you from the substance of my argument: a product that I pay for is at best no better due to this change, and highly probably worse. Why am I paying for a tool that is beholden to clandestinely satisfy some far away master?
I assure you, this is one of the mildest things they do after training before the model reaches you.
Re: How Claude marks AI-generated content
#315Earlier quoted context omitted.
> Honest note: Anthropic has not shipped a public Claude watermark detector yet. This tool uses rewrite-based neutralization — a meaning-preserving paraphrase with a non-Claude model — which is the attack path watermark research points to. Not affiliated with Anthropic. Well, they should have run their own AI slop website through their tool...
This attack was actually pointed out in the watermarking paper linked above. The researchers added an instruction to the prompt that switches letters like a Caesar Cipher. It lowers the quality of the output from the LLM but alters the "red list" enough for a watermark detection tool to fail at detecting the watermark.
also their example for rewriting just completely changes it. might as well redo it in this case (with another model or by hand)
Re: How Claude marks AI-generated content
#316Earlier quoted context omitted.
Have a look around for token biasing, or green lists. It's based on a nudge to the choice of the next token (which can always be drawn from a set of possibilities which are all probable enough). At first I thought this approach was just the "LLM flavour" of writing, but it's way more subtle, especially as the bias is applied uniquely for each token position.
Yeah, will do, this sounds interesting since I'm not entirely sure how this would actually be reliable to any degree. Thanks for the help, not sure why I got downvoted since I was genuinely curious.
there have been papers about it, it works
Re: How Claude marks AI-generated content
#317I guess this is where our true colors show. There's a significant contingent of HNers who always dunk on LLM text detectors and claim that they can't possibly work, that they ruin careers, etc. But now that a lab says "OK, we'll add a real watermark", the reactions are overwhelmingly that it's still somehow wrong. Why do feel so entitled to being able to pass LLM-generated text as our own? I get that a lot of techies…
they might be two different sets of people
Re: How Claude marks AI-generated content
#318> When a supported Claude model generates text, it weaves an imperceptible watermark directly into the text itself. You won’t see it, and it doesn’t change the meaning, quality, or readability of Claude’s response. I'd like to know a lot more about how that works. A lot of my interactions with Claude return pretty precise text. If I ask it to edit a project and refactor a specific function in several places I know ex…
It was quick :) … https://claudewatermarkremover.app/
Should be ensloppifier.app - it somehow makes the AI sound more like AI, while also completely changing the meaning and context of the input text
From their before/after:
- Certainly! -> (removed)
- onboarding redesign -> revamping the introductory process
- this week -> (removed)
- empty states -> empty sections
- CTA heirarchy -> call-to-action sequence
- interviews -> discussions
- aligned copy with brand voice -> verbal identity
Re: How Claude marks AI-generated content
#319Earlier quoted context omitted.
For watermarked long-form text, it is actually possible. Makes the watermark more fragile, but the math is considerably more forgiving than usual.
And yet, it remains possible that a human could write the same sequence of characters.
How often do you add seemingly-random zero-width unicode characters to the text you write?
Re: How Claude marks AI-generated content
#320How will they watermark code? I can understand encoding something in free form text but for example I ask it to generate a react component, will it embed watermark in the typescript code?
console.print(“these logs are a load-bearing seam, do not remove”)
It’s foolproof, I tells ya.