OpenAI Afraid to Release ChatGPT Detection Tool That Might Piss Off Cheaters:
ChatGPT maker OpenAI has new search and voice features on the way, but it also has a tool at its disposal that's reportedly pretty good at catching all those AI-generated fake articles you see on the internet nowadays. The company has been sitting on it for nearly two years, and all it would have to do is turn it on. All the same, the Sam Altman-led company is still contemplating whether to release it as doing so might anger OpenAI's biggest fans.
This isn't that defunct AI detection algorithm the company released in 2023, but something much more accurate. OpenAI is hesitant to release this AI-detection tool, according to a report from the Wall Street Journal on Sunday based on some anonymous sources from inside the company. The program is effectively an AI watermarking system that imprints AI-generated text with certain patterns its tool can detect. Like other AI detectors, OpenAI's system would score a document with a percentage of how likely it was created with ChatGPT.
OpenAI confirmed this tool exists in an update to a May blog post posted Sunday. The program is reportedly 99.9% effective based on internal documents, according to the WSJ. This would be far better than the stated effectiveness of other AI detection software developed over the past two years. The company claimed that while it's good against local tamping, it can be circumvented by translating it and retranslating with something like Google Translate or rewording it using another AI generator. OpenAi also said those wishing to circumvent the tool could "insert a special character in between every word and then deleting that character."
Internal proponents of the program say it will do a lot to help teachers figure out when their students have handed in AI-generated homework. The company reportedly sat on this program for years over concerns that close to a third of its user base wouldn't like it. In an email statement, an OpenAI spokesperson said:
"The text watermarking method we're developing is technically promising, but has important risks we're weighing while we research alternatives, including susceptibility to circumvention by bad actors and the potential to disproportionately impact groups like non-English speakers. We believe the deliberate approach we've taken is necessary given the complexities involved and its likely impact on the broader ecosystem beyond OpenAI."
The other problem for OpenAI is the concern that if it releases its tool broadly enough, somebody could decipher OpenAI's watermarking technique. There is also an issue that it might be biased against non-native English speakers, as we've seen with other AI detectors.
(Score: 5, Interesting) by bzipitidoo on Wednesday August 07, @01:47PM (5 children)
I posted my essay that copyright thinking twists art to a forum on literature. Got strongly negative responses, with at least two posters making the accusation that the essay was so bad it must have been written by AI. I don't think they seriously believed their own accusations, I think they didn't like what the essay said, and were just looking for a cheap way to smear it. A quick scan of that forum shows that "written by AI" is their new go-to insult.
I suppose I should have expected that literature fans would be almost as anti-copying as artists. There's no reasoning with artists or publishers on matters of copyright. Nor, apparently, their fans. It will have to be continued technological advance and the rest of us who eventually make them face the new reality.
(Score: 5, Insightful) by Freeman on Wednesday August 07, @02:57PM (2 children)
Even were I a competent artist/writer/etc. I would still be all for a much more sane copyright term. There's absolutely no reason for content from over 70 years ago to still be under copyright. Okay, no reason that was intended! The reason now is for large corporations to suck up the profits from X popular author that died many decades ago. This goes especially so for software, where the item has been abandoned for a couple of decades or more and still won't be out of copyright for a several more decades. Be the time which there likely won't be a functional original.
Joshua 1:9 "Be strong and of a good courage; be not afraid, neither be thou dismayed: for the Lord thy God is with thee"
(Score: 4, Insightful) by Thexalon on Wednesday August 07, @05:55PM (1 child)
Slight correction here:
Authors usually retain their copyrights on books, so it's not the megacorps pulling in the cash but their descendants once you get into the "50 years after they're dead" clause.
The stuff where the megacorps are really raking it in are either works for hire (most stuff on video with a large production team) or where they gained the copyrights through contract (most musical recordings).
And yes, you are right that newer creators benefit from shorter copyright. After all, why should Disney cultivate the next big thing when they can just milk their existing portfolio forever?
The only thing that stops a bad guy with a compiler is a good guy with a compiler.
(Score: 2) by Freeman on Thursday August 08, @02:53PM
You are correct that there is some resistance where individual authors are concerned. I would posit that it's likely only those that have hit it big and the corporations that are the big hurdle there. Though, there could be a lot more authors that hope to hit it big and dream about the big payday than I think.
Joshua 1:9 "Be strong and of a good courage; be not afraid, neither be thou dismayed: for the Lord thy God is with thee"
(Score: 2) by Ox0000 on Wednesday August 07, @08:13PM (1 child)
You did not answer the accusation... was it? ;P
(Score: 3, Informative) by bzipitidoo on Wednesday August 07, @11:06PM
I told them that I wrote it, and that I am fully human, and I denied that it was written by an AI. Didn't even use AI just to help, and I said that too.
My denial was then accused of also being written by an AI that was lying. And taken as evidence that AI's must be getting pretty smart, able to tell a lie. Sometimes, you just can't win.
(Score: 5, Informative) by owl on Wednesday August 07, @03:06PM
The old adage:
Applies equally here, only translated to:
(Score: 4, Insightful) by Ken_g6 on Wednesday August 07, @05:24PM
Even assuming OpenAI shares this technology with Google and Facebook, I could download a copy of Mistral, or an old LLaMA, and write AI-generated text without the watermark. Even if legislation mandated this watermarking system, these old systems will be somewhere on the Internet forever.
(Score: 4, Interesting) by looorg on Wednesday August 07, @06:10PM (1 child)
I'm not really to concerned about the cheaters. Fuck them. I'm more concerned about people that will be falsely accused of cheating but will basically have no way of defending themselves. We already have issue with all the anti-plagiarism software. If that thing tags you then you are fucked. There is no way of proving that you didn't read that other book, paper or whatever. Cause apparently in fantasy-land there is an infinite number of ways to express things. So no two text explaining the same thing should ever look the same. Right. So these days you have to run these on yourself to see if you look like someone else, if you do the cite them instead. It's so wrong in so many ways.
That said I'm wondering how they will watermark text. Hidden characters? Hashes? Patterns or phrases? Spelling errors? All of these as far as I can think of now can usually be fairly easily removed or corrected for. If one knows they exist. The once described often seem to require a text of a certain minimum size, at least 1000 words, usually around 500ish words per page (but you could probably half that if you have some graphs or uses a lot of paragraphs) if you write normally with font-size and spacing and such. Then assigning values to certain words and phrases. The look for them. But it's still probabilities. Which will leave some people falsely accused and fucked by the machine, not in any pleasant way.
So to bad for you if you have a writing style that the AI monstrosity deems to feel and look like AI. Better not get a job that requires any kind of writing.
(Score: 3, Interesting) by VLM on Thursday August 08, @06:23PM
It might be simpler than you think. Back in school when kids would paraphrase paper encyclopedia articles instead of writing their own essays or copy stuff out of Cliffs Notes (does that still exist?) the prof/teacher/TA would simply say "hey lets talk about your essay". I found those conversations pretty easy because I wrote my essay. I would imagine if you copied someone else (including the process of copying a corporation's article, or nowadays AI) that would be a rather disjointed conversation.
"So how did you come up with your second paragraph?" If I just wrote the thing, it's pretty easy to crap out two to four lines instantly that sound plausible. Or if you copied some article or AI output, uh, what was "my" second paragraph about again, um?
I suspect the main way to cheat with AI would be asking the AI for something like old fashioned Cliffs Notes where you don't cut and paste anything, you're just given the hint. If you're assigned some diversity nonsense essay like discuss which of William Shakespeare's characters were lesbians, that's a LOT of work if you have to read ALL of his works and think about all of them, but an AI could tell you "well obviously Portia is widely believed to be an alphabet person" and then you can follow up on your own. I'd have to think about Portia ... probably a lot of projecting going on if someone claims she's "out", someone's "out" if Portia is being claimed to be "out" but its not Portia LOL, however maybe as a spectrum thing she was a little ... overly friendly. You'd have to read Merchant of Venice to make your own mind up. I've read some real dreck about "oh she's very resourceful so obviously that proves she's a lesbian" which sounds pretty LOL to me. Anyway, that's how you use AI to cheat without being caught, use it like a sounding board or proof reader or magic outline writer.
I have played around with AI "a lot" and I've noticed language and idiom is incredibly stilted and monocultural. I probably read like an upper-midwestern upper middle class white dude who's had a beer after getting out of work early, because I am, but after you read AI for long enough, you'll recognize its weird style, instantly. Some of its grammar choices are very repetitive and obvious. Its possible if AI lasts long enough, everyone who learns English worldwide will sound exactly identical; I hope not.
(Score: 3, Funny) by Ox0000 on Wednesday August 07, @08:11PM
Let me guess, it adds the sentence "This content was generated by ChatGPT" as watermarking, amirite?