Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Undetectable Watermarks for Language Models

eprint.iacr.org

Post reply on HN
Print view View original

21–30 of 71 posts

Re: Undetectable Watermarks for Language Models

#21

Post by alvah »

dbgev wrote: post #15
chaxor wrote: post #11

Earlier quoted context omitted.

The only real way to block specifically openAI generated content (or some other online generated LLM content) is for the company itself to store all of its outputs and compare to that database, like shingling / LSH for plagiarism detection. Other LLMs (local) are completely impossible to block, as it's a constant chase. Any system that tries to estimate the distribution of e.g. a specific LLM that performs beam searc…

It doesn't need to stop the determined users, thete's value in just filtering out articles that are not even proofread by humans. It's going to be a constant cat and mouse game, I agree, but for now it just need to catch the dead mice.

The phrase "As an AI language model" could be useful here.

Re: Undetectable Watermarks for Language Models

#22

Post by dbgev »

alvah wrote: post #21
dbgev wrote: post #15

Earlier quoted context omitted.

It doesn't need to stop the determined users, thete's value in just filtering out articles that are not even proofread by humans. It's going to be a constant cat and mouse game, I agree, but for now it just need to catch the dead mice.

The phrase "As an AI language model" could be useful here.

Is accusing people of being an AI model going to become the new Godwin's law?

Re: Undetectable Watermarks for Language Models

#24

Post by mach1ne »

EMM_386 wrote: post #16

There are obviously ways it can be "watermarked" easily, put some zero-width unicode characters in the output and you'll notice right away when it's copy and pasted. But clearly, that can be stripped out easily by anyone who knows it's there. This process, too, would seem to be easily reversible. Just have it run through another model and tell it to slightly reword it or rephrase it. I don't think there is a technica…

I, for one, do believe watermarking solutions exist. One thing you cannot escape with LLMs is content meaning.

As a simple example, the secret watermark could be hidden in the embeddings of the sequence of words. To make the watermark more robust against rephrasings, it could be hidden in the meaning of sentences or paragraphs.

At the minimum, I think this could be possible.

Re: Undetectable Watermarks for Language Models

#26

Post by speedgoose »

m3kw9 wrote: post #5
valine wrote: post #2

What’s to stop someone from taking the watermarked output and randomizing the distribution by feeding it through their latest LLaMA variant? These watermarks will only be useful for catching novice LLM users.

Yeah soon there will be models small enough to run even on phones to reword things slightly differently. If not, an app will do it

I already have one running on my phone: https://mlc.ai/mlc-llm/

Re: Undetectable Watermarks for Language Models

#29

Post by schrodinger »

mach1ne wrote: post #24
EMM_386 wrote: post #16

There are obviously ways it can be "watermarked" easily, put some zero-width unicode characters in the output and you'll notice right away when it's copy and pasted. But clearly, that can be stripped out easily by anyone who knows it's there. This process, too, would seem to be easily reversible. Just have it run through another model and tell it to slightly reword it or rephrase it. I don't think there is a technica…

I, for one, do believe watermarking solutions exist. One thing you cannot escape with LLMs is content meaning. As a simple example, the secret watermark could be hidden in the embeddings of the sequence of words. To make the watermark more robust against rephrasings, it could be hidden in the meaning of sentences or paragraphs. At the minimum, I think this could be possible.

How about you output in Spanish, then google translate to English?

Re: Undetectable Watermarks for Language Models

#30

Post by bryanrasmussen »

Why would you want to watermark your content - generally watermarks are used to provide legal proof of provenance which can be important when suing someone for stealing your content but since machine learning outputs cannot be copyrighted this use is not important.
Post reply on HN