You still have a series of words with first letters spelling out CHATGPT.
1 Like
Yeeeeeah… but that way lies the Bible Code and madness.
4 Likes
Except that it doesn’t reliably “easily identify low-effort uses”, and once introduced will immediately induce low-effort circumventions. If one dials the detector gain it starts generating false positives, and that causes real harm to real people, as opposed to making some SEO optimizers job trivially more difficult.
Language is simply too malleable a medium for these kinds of counter measures to work. Ever.
OpenAI made the right decision by not deluding the public into believing they could work. Other companies have chosen another route, and the results have been entirely as expected.
Here is but one experiment on the topic:
No public AI text detector we tested scored better than random chance. Results were very unstable, with small changes to input text flipping detections in both directions. LLMs also failed to reliably detect LLM output in our tests.
Now, these are external tools, not internal watermarking, but the lessons are the same: this isn’t a solvable problem and it never will be.
1 Like
Who cares if LLM-generated text can be detected algorithmically. More by the day I’ll land on a website and my human intelligence gets the feeling that it was made by an AI product. Sure, my feelings won’t be 100% accurate. But eventually the situation isn’t going to be “hey i think this might be machine made,” but rather, “wow, this is actual human writing, how refreshing!”
Oftentimes i dive into these BB posts without noting the author. And from time to time, I’m halfway through one and i think, “ha, this is a beschizza piece,” and i scroll up to check and I’m right every single time. Because humans can construct original things. Machines can’t do that.
Someday? Won’t say it’s impossible, but on the data-information-knowledge-wisdom hierarchy these text outputs are still on the data level. At best they can synthesize rudimentary information.
1 Like
A major problem is all the LLM generated slop (“slop” is the technical term for LLM output) is being fed back into LLMs as new training data. Unable to tell valuable content from slop, they are rapidly devolving themselves as they eat their own tails.
And while LLMs rendering themselves useless sounds like a good thing, the problem is it still leaves the slop inextricably mixed in with the extant human content.
3 Likes
That’s a huge problem now, though, as there are no reliable ways of detecting LLM output. Teachers are running student papers through software that promises to detect LLM writing which a) doesn’t do that, often, and b) is full of false positives. As for SEO scams, if Google (et al) detect that every page of a website has these indicators, they can pretty safely downrank it. Is watermarking a perfect solution? Obviously not, but it is a solution that works better than what currently exists.
1 Like
This topic was automatically closed after 5 days. New replies are no longer allowed.
