The most prestigious law school admissions discussion board in the world.
|
 |
 
|
New Claude models to include watermarks inside text - link
Poast new message in this thread
Date: August 10th, 2026 5:29 PM Author: provocative cream pozpig property https://x.com/M1Astra/status/2086898041882030353
Does this mean teachers, judges, etc will be able to know you used Claude AI? Seems like a gift for ChatGPT if so. |
Date: August 10th, 2026 5:31 PM Author: opaque famous landscape painting background story Open AI already looked into doing the same thing and decided there were major drawbacks (http://www.autoadmit.com/thread.php?thread_id=5891530&forum_id=2…id#50058695) |
 |
Date: August 10th, 2026 5:37 PM Author: Bronze sanctuary It seems useless anyway since you'd only have to change a few words and the only value is AI companies and governments acting like they're regulating AI transparency. (http://www.autoadmit.com/thread.php?thread_id=5891530&forum_id=2…id#50058710) |
|
Date: August 11th, 2026 1:45 PM Author: Gold hairraiser boltzmann Changing a few words, or even perhaps the entire thing, doesn’t necessarily get rid of the watermark. (http://www.autoadmit.com/thread.php?thread_id=5891530&forum_id=2…id#50060923) |
|
Date: August 11th, 2026 1:48 PM Author: opaque famous landscape painting background story Yes it does, because the entire "watermark" is based on biasing the choice of certain tokens over others. It's extremely easy to attack. (http://www.autoadmit.com/thread.php?thread_id=5891530&forum_id=2…id#50060928) |
|
Date: August 11th, 2026 1:50 PM Author: Bronze sanctuary The watermark is encoded into the plain text, so yes it would. Obviously they can make it difficult but that puts more constraint on what's possible to generate. (http://www.autoadmit.com/thread.php?thread_id=5891530&forum_id=2…id#50060933) |
|
Date: August 11th, 2026 1:54 PM Author: Wonderful ultramarine hospital Oh wow an LLM that encodes my text with hidden messages I can't parse *signs up for a paid plan* (http://www.autoadmit.com/thread.php?thread_id=5891530&forum_id=2…id#50060944) |
|
Date: August 10th, 2026 5:33 PM Author: Glittery sadistic skinny woman goal in life idgi. cant you just be like claude write me a 500 wd essay, then print it out and retype it into word? So you're just adding one step.
tbf even adding that one step might actually cut down significantly on ai plaigiarism. idk. people react funny to things |
|
Date: August 10th, 2026 5:34 PM Author: opaque famous landscape painting background story No the "watermark" isn't in metadata or characters. It's a token bias that causes certain tokens to be selected over others in a way that doesn't change the overall meaning of the text, but it can be recovered that it was made by claude. You can still destroy it by rewriting pieces of the output or running it through a translator and back.
One of the reasons Open AI decided not to do this is they thought it could cause a DITHPARATE IMPACT to ESL users |
|
Date: August 10th, 2026 5:36 PM Author: Stubborn theater place of business it's done via semantic patterns woven into the textual language itself (http://www.autoadmit.com/thread.php?thread_id=5891530&forum_id=2…id#50058707) |
|
Date: August 10th, 2026 5:38 PM Author: Glittery sadistic skinny woman goal in life wowza
like how shakespear supposedly snuck his name into Psalm 46 |
|
Date: August 10th, 2026 5:42 PM Author: Cerebral dull national Yeah I worked on something like this in college for a corporate espionage thing (http://www.autoadmit.com/thread.php?thread_id=5891530&forum_id=2…id#50058720) |
|
Date: August 10th, 2026 9:58 PM Author: Excitant Flickering Candlestick Maker huh. so do you break the pattern by randomly changing every fifth or tenth word or whatever? (http://www.autoadmit.com/thread.php?thread_id=5891530&forum_id=2…id#50059650) |
|
Date: August 10th, 2026 10:01 PM Author: opaque famous landscape painting background story They statistically bias certain tokens to enough of an extent that it leaves a detectable pattern, but they aren't releasing the exact pattern afaict. I would just put Claude's output into gpt and tell it to rewrite it in different words while maintaining the meaning. (http://www.autoadmit.com/thread.php?thread_id=5891530&forum_id=2…id#50059663) |
|
Date: August 10th, 2026 10:06 PM Author: Excitant Flickering Candlestick Maker ty cr. that's my reaction to this tactic also - just requires that the user do an extra step, and towards what end? i don't understand why they would do this (http://www.autoadmit.com/thread.php?thread_id=5891530&forum_id=2…id#50059691) |
|
Date: August 10th, 2026 10:14 PM Author: opaque famous landscape painting background story Open AI researched the same thing, and they decided not to do it, and one of their reasons was that it is really easy to break. Apparently also the more you try to harden it the more it has an affect on the quality of the output itself. (http://www.autoadmit.com/thread.php?thread_id=5891530&forum_id=2…id#50059716) |
|
Date: August 10th, 2026 10:23 PM Author: Excitant Flickering Candlestick Maker huh cr. the way i'm thinking about it is forcing content into iambic pentameter or whatever. sometimes the rule produces beautiful content, and a constraint may help to shape human creativity. but other times it just results in a stretch/cheat/weird content (http://www.autoadmit.com/thread.php?thread_id=5891530&forum_id=2…id#50059744) |
|
Date: August 10th, 2026 10:31 PM Author: opaque famous landscape painting background story Yeah that does seem like a good analogy. Like meter is nice to have for poetry, but at the same time it is constraining your generative creative capability, forcing you to make choices to fit the meter. Where here it probably comes down to if you bias too many tokens then you get a shittier output. More tokens you bias, more resilient the watermark, but then you risk forcing worse choices to fit the algorithm. (http://www.autoadmit.com/thread.php?thread_id=5891530&forum_id=2…id#50059775) |
|
Date: August 11th, 2026 8:11 PM Author: Excitant Flickering Candlestick Maker (http://www.autoadmit.com/thread.php?thread_id=5891530&forum_id=2…id#50062187) |
|
Date: August 10th, 2026 5:33 PM Author: bisexual stage factory reset button How would this work? If it's part of the text, the user can just delete it--assuming they can be bothered to look through the text to find it. (http://www.autoadmit.com/thread.php?thread_id=5891530&forum_id=2…id#50058700) |
|
Date: August 10th, 2026 5:36 PM Author: Multi-colored tanning salon "it will travel with the text when it's copied and pasted elsewhere, and may persist through some editing."
they know it can just be deleted, but many people are dumb enough to do a blatant copypasta |
|
Date: August 10th, 2026 5:40 PM Author: Cerebral dull national You could watermark AI text by using a secret key to bias otherwise natural word choices.
At selected points, a keyed PRNG decides which of several semantically equivalent choices is preferred. Those choices encode bits, and a verifier with the same key checks whether the observed choices correlate with the expected pattern.
For example:
"however" = 0, "but" = 1 "large" = 0, "big" = 1 "because" = 0, "since" = 1
If the secret pattern is:
1 0 1 1 0 0 1...
the model subtly favors the corresponding choices. To a reader, the prose still looks completely normal.
A more sophisticated version would not use obvious synonym pairs. Instead, candidate next tokens could be secretly divided into two groups with something like:
bit(token) = HMAC(K, context || token) mod 2
To encode a 0, slightly favor tokens from group 0; to encode a 1, favor group 1. The result is a keyed statistical signature rather than an obvious hidden message.
You also would not need every bit to survive. Add redundancy with an error-correcting code, parity bits, and repeated/interleaved copies of the watermark throughout the text. Then even if large sections are deleted, reordered, or rewritten, the verifier can recover enough surviving signal to identify the watermark statistically.
For example, across 500 surviving tokens, if 68 percent match the secret expected groups instead of the roughly 50 percent expected by chance, the verifier can assign a confidence score that the text came from the watermarked generator.
|
|
Date: August 10th, 2026 5:45 PM Author: opaque famous landscape painting background story Yeah its retarded though because you can technically destroy it just by changing one token, because that can alter the watermark state for subsequent positions. Theoretically, depending on the specific strategy they are using for biasing it, and which token you pick. But the more robust they try to make the watermarking, the more constrained and janky the writing and output capability becomes. (http://www.autoadmit.com/thread.php?thread_id=5891530&forum_id=2…id#50058734) |
|
Date: August 10th, 2026 11:56 PM Author: bisexual stage factory reset button Interesting. It's going to get people into trouble when it looks like the text might be from AI but it's not totally clear. (http://www.autoadmit.com/thread.php?thread_id=5891530&forum_id=2…id#50060025) |
|
Date: August 10th, 2026 5:43 PM Author: Bronze sanctuary A pattern is encrypyed into the text. Think something like making every word occurring in a prime number sequence starting with an advancing vowel, but obviously more complicated. An AI could write anything under those rules without changing the content or quality that much. But yeah, obviously anything encrypted into plain text can be destroyed by simply changing a couple words. (http://www.autoadmit.com/thread.php?thread_id=5891530&forum_id=2…id#50058724) |
|
Date: August 10th, 2026 9:58 PM Author: Excitant Flickering Candlestick Maker (http://www.autoadmit.com/thread.php?thread_id=5891530&forum_id=2…id#50059652) |
|
Date: August 10th, 2026 5:39 PM Author: Aphrodisiac misanthropic base juggernaut taking a screenshot then telling chatgpt to type the text from it is an obvious workaround (http://www.autoadmit.com/thread.php?thread_id=5891530&forum_id=2…id#50058714) |
|
Date: August 10th, 2026 5:46 PM Author: opaque famous landscape painting background story Huh? How would that do anything? You can just copy and paste it into gpt and tell it to rewrite it in different words and it will destroy the watermark. (http://www.autoadmit.com/thread.php?thread_id=5891530&forum_id=2…id#50058739) |
|
Date: August 10th, 2026 5:44 PM Author: Bronze sanctuary AI writing "I want to kill myself too" next to your 4pt white text watermark. (http://www.autoadmit.com/thread.php?thread_id=5891530&forum_id=2…id#50058727) |
|
Date: August 10th, 2026 5:49 PM Author: brass philosopher-king kitty cat "I'm gay"
- This text was generated using Claude Fable 5, a "Mythos-level" frontier AI model by Anthropic. |
|
Date: August 10th, 2026 5:52 PM Author: Stubborn theater place of business (http://www.autoadmit.com/thread.php?thread_id=5891530&forum_id=2…id#50058754) |
|
Date: August 10th, 2026 5:52 PM Author: Boyish angry market (http://www.autoadmit.com/thread.php?thread_id=5891530&forum_id=2…id#50058756) |
|
Date: August 10th, 2026 5:58 PM Author: Glittery sadistic skinny woman goal in life (http://www.autoadmit.com/thread.php?thread_id=5891530&forum_id=2…id#50058771) |
|
Date: August 10th, 2026 5:59 PM Author: curious casino love of her life (http://www.autoadmit.com/thread.php?thread_id=5891530&forum_id=2…id#50058776) |
|
Date: August 10th, 2026 6:02 PM Author: opaque famous landscape painting background story (http://www.autoadmit.com/thread.php?thread_id=5891530&forum_id=2…id#50058782) |
|
Date: August 11th, 2026 2:09 PM Author: frisky psychic (http://www.autoadmit.com/thread.php?thread_id=5891530&forum_id=2…id#50060995) |
|
Date: August 10th, 2026 6:03 PM Author: Internet-worthy bipolar plaza Write your own words you weird fags (http://www.autoadmit.com/thread.php?thread_id=5891530&forum_id=2…id#50058785) |
|
Date: August 11th, 2026 1:00 PM Author: Rambunctious School Cafeteria Cuck "It's too hard to think." (xo po (http://www.autoadmit.com/thread.php?thread_id=5891530&forum_id=2…id#50060818) |
|
Date: August 11th, 2026 1:53 PM Author: Bistre Duck-like Box Office Faggotry (http://www.autoadmit.com/thread.php?thread_id=5891530&forum_id=2…id#50060943) |
|
Date: August 10th, 2026 10:41 PM Author: Olive tank hall it's an acrostic that spells out 'I AM GAY' (http://www.autoadmit.com/thread.php?thread_id=5891530&forum_id=2…id#50059794) |
|
Date: August 10th, 2026 11:37 PM Author: Cerebral dull national (http://www.autoadmit.com/thread.php?thread_id=5891530&forum_id=2…id#50059957) |
|
Date: August 11th, 2026 1:12 PM Author: brass philosopher-king kitty cat (http://www.autoadmit.com/thread.php?thread_id=5891530&forum_id=2…id#50060852) |
|
Date: August 11th, 2026 1:46 PM Author: Gold hairraiser boltzmann Next step. Release higher cost “watermark free” subscription tier. (http://www.autoadmit.com/thread.php?thread_id=5891530&forum_id=2…id#50060926) |
|
Date: August 11th, 2026 1:55 PM Author: Bistre Duck-like Box Office Faggotry can some1 pls answer a q. why, tf, is this subject worthy of 40 poas and the reems of screed included within each? this seems like the most trivial shit one could possibly conceive of which? (http://www.autoadmit.com/thread.php?thread_id=5891530&forum_id=2…id#50060945) |
|
Date: August 11th, 2026 2:01 PM Author: Bronze sanctuary It's funny when people complain about a thread they don't enjoy getting more traction then they think it deserves, thus increasing the poast count they're complaining about. (http://www.autoadmit.com/thread.php?thread_id=5891530&forum_id=2…id#50060967) |
|
Date: August 11th, 2026 2:04 PM Author: Bistre Duck-like Box Office Faggotry no bro it's not that i don't enjoy this thread i just don't understand why this topic is worthy of literally anything longer than like maybe 0 seconds of discussion? can you help me understand why this is important or even worthy of a second thought? since you're a super smart and good looking dude, i expect i'll be all in on watermarks or not in if that's what you want me to think after you explain (http://www.autoadmit.com/thread.php?thread_id=5891530&forum_id=2…id#50060975) |
|
Date: August 11th, 2026 2:10 PM Author: Bronze sanctuary It's interesting to me that they're encoding a watermark into plain text and that there are efforts to make AI identifiable at all.
Retardless, this isn't some bait thread with a misleading title. If it doesn't interest you don't click on it. And especially don't click it again when the link turns blue to read updates. The frontpage is 90% NSAM spam, it's not like this thread getting bumped is competing with quality content. |
|
Date: August 11th, 2026 2:17 PM Author: Bistre Duck-like Box Office Faggotry ok because you are so smart and sexy and i bet you're one of those hot nerds whose like pretty muscular and i like you (like that), i read up on this a bit. i am now interested in it given EU GDPR-like regulation is the driving force. once again, our poor and slightly fruity European cousins have saved the day and shown the cajones to push back against the Rise of the Machines (http://www.autoadmit.com/thread.php?thread_id=5891530&forum_id=2…id#50061016) |
|
Date: August 11th, 2026 2:34 PM Author: Bronze sanctuary Idk why you're so irate and implying I think this thread should have broad interest. It is a niche topic, and clearly labeled, if it doesn't interest you just don't click? Why do you care others are interested? (http://www.autoadmit.com/thread.php?thread_id=5891530&forum_id=2…id#50061062) |