Students cheating with LLMs show more a weakness of a system than anything else. Rich folks had other people make assignment for them for ages, now this got democratized. The educational institutions are mostly disinterested in really doing what they should do to teach critical thinking and understanding if students really understood. To check the home made assignments is a small part of that (in theory). It's not LLMs, is terrible schools.
I studied CS and Philosophy in college. The philosophy department at the time was struggling to get people to declare so they started a campaign with the slogan “Thinking of a major? Major in thinking”. Which I’ve always thought was both clever and accurate. Such departments become more valuable as the harder skills are more automatable. I doubt that will translate into more philosophy majors but one can dream.
They can when they can. Sometimes they just won't like a kid or be tired or just wrong and falsely accuse them. There's been plenty already. I had a kid in my class during some years that would brag to other kids that his mom wrote all his longer, non-math homeworks. You're only putting yourself back by cheating, when time comes for proper exams for university placement you won't have it available and you won't take anyone else's place. Most places nowadays won't even fail kids from passing to the next grade, so it doesn't even matter.
The examples they give all look like valid uses of different Non-breaking spaces, with width hints for their use/location, this might be a little overzealous if written by a human but perhaps not for a machine.
Yeah, it's probably not an intentional watermark, just something the model has been trained to do. Maybe some professionally written news articles already use them for the same purpose?
Still hope HN adds a filter to block any comment with those characters in it :)
It is very easy to filter those out from the output of GPT, though, using basic UNIX utilities. In fact, many methods don't survive reformatting or copy-pasting, not requiring filtering at all.
It is a very basic watermark technique (text steganography) if it indeed is supposed to be one.
A more advanced one would be a linguistic (grammar-based) one, but I am not going to give any more ideas. :D
I mean, sure, these characters could be used to help estimate the likelihood text was generated (because human writers might be less likely to add proper non-breaking spaces), but I doubt these are watermarks.
Ideally this watermarks are much more subtle. For example they put watermark through different distribution of word sequence which are difficult to remove or identify
Readit News
It's not watermarks, it's just scraped typography.
https://en.wikipedia.org/wiki/Non-breaking_space
Different display apps may "display them identically" but others and typesetters/printing apps might not.
Still hope HN adds a filter to block any comment with those characters in it :)
It is a very basic watermark technique (text steganography) if it indeed is supposed to be one.
A more advanced one would be a linguistic (grammar-based) one, but I am not going to give any more ideas. :D
I mean, sure, these characters could be used to help estimate the likelihood text was generated (because human writers might be less likely to add proper non-breaking spaces), but I doubt these are watermarks.
A shocking title which brought this to the HN frontpage, but then it does not hold. The characters all look legit in the position they are used.
Google do it with Gemini tho:
https://lifearchitect.ai/watermarking/