OpenAI has a 99.9% accurate ChatGPT AI text detector, but won't release it. (the-decoder.com)
from ModerateImprovement@sh.itjust.works to technology@lemmy.world on 04 Aug 2024 14:23
https://sh.itjust.works/post/23204457

#technology

threaded - newest

Alphane_Moon@lemmy.world on 04 Aug 2024 14:38 next collapse

Given a sufficient amount of text, the method is said to be 99.9 percent effective.

If that’s really the case, they should release some benchmarks. I am skeptical. Promising the world is a key component of their “business model”.

NegativeInf@lemmy.world on 04 Aug 2024 15:01 next collapse

What is a sufficient amount? Most comments are short af.

technocrit@lemmy.dbzer0.com on 04 Aug 2024 15:01 next collapse

I don’t think these grifters know what a benchmark is.

MagicShel@programming.dev on 04 Aug 2024 15:10 collapse

I think given enough output I could probably detect it that accurately as well. ChatGPT has a particular voice and the longer it goes, the more that voice comes out.

hendrik@palaver.p3x.de on 04 Aug 2024 14:47 next collapse

That's a bad article. What are they reluctant about? Releasing that detector, or applying watermarks to the generated texts? Do they do that already or doesn't it apply to text generated until then? And how would that affect anything else?

Whats with the error rate? Shouldn't that be near 100% for watermarks? And 0 false positives? What's really holding them back? Is pupils not turning in ChatGPT homework anymore cutting into their business model?

I mean all the major AI companies promised to do AI ethically. Now they don't want the one thing that would solve half the issues people are having with that technology. Kind of fits with OpenAI 🤔

drmoose@lemmy.world on 05 Aug 2024 08:05 collapse

They can’t release anything as watermarks can be reverse engineered and people would just wise up and tumble the outputs.

Weirdly, not releasing this tool publicly might be the smartest bet here as all of these bot farms and idiots just blindly use chatgpt outputs without any tumbling or safety.

hendrik@palaver.p3x.de on 05 Aug 2024 09:28 collapse

The issue with that is: Releasing nothing is even worse than releasing something that could be circumvented. I don't see this as a valid argument.

I'm not an expert on text watermarking and how that degrades output. But if they want some stealthy solution that isn't known to the public... Maybe they could attach two watermarks. A simple one that is known to everyone, and an additional, secret one only they know about. It'd be similar to what we do with bank notes. There are some characteristics everyone knows and can use to judge if it's fake money. And they have some additional secret markings in banknotes that only the central bank knows about.

I'm pretty sure a similar thing could be done here. Maybe not for a 280 character tweet. But certainly for other use-cases with longer texts. And in case it has a 0% false positive rate, every match helps someone. Even if it's circumventable. I think even a non-perfect solution that helps several thousands of people is better than helping no-one.

Pika@sh.itjust.works on 05 Aug 2024 13:31 collapse

I agree with not releasing it, but I do find that it defeats the purpose talking about it because if you have it but aren’t sharing if what’s the point of having it

hendrik@palaver.p3x.de on 05 Aug 2024 15:15 collapse

I think we're missing half the story. Because I also fail so see a point in doing it like they do.

FlyingSquid@lemmy.world on 04 Aug 2024 14:54 next collapse

I wonder if this means they’ve discovered a serious flaw that they don’t know how to fix yet?

MagicShel@programming.dev on 04 Aug 2024 15:12 next collapse

The flaw is in the training to make it corporate friendly. Everything it says eventually sounds like a sexual harassment training video, regardless of subject.

ArbitraryValue@sh.itjust.works on 04 Aug 2024 15:21 collapse

I think the more like explanation is that being able to filter out AI-generated text gives them an advantage over their competitors at obtaining more training data.

expatriado@lemmy.world on 04 Aug 2024 14:54 next collapse

shhh, my professor may use it

PenisDuckCuck9001@lemmynsfw.com on 04 Aug 2024 20:44 next collapse

My unpopular opinion is when they’re assigning well beyond 40 hours per week of homework, cheating is no longer unethical. Employers want universities to get students used to working long hours.

amanda@aggregatet.org on 05 Aug 2024 07:24 collapse

I agree, and I teach. A huge part of learning is having the time to experiment and process what you’ve learnt. However, doing that in a way that can be controlled, examined, etc, is very difficult so many institutions opt for tons of homework etc.

amanda@aggregatet.org on 05 Aug 2024 07:25 collapse

If the assignment is so easy ChatGPT can do it, it’s too easy.

tinfoilhat@lemmy.ml on 04 Aug 2024 14:59 next collapse

I call bullshit.

superkret@feddit.org on 04 Aug 2024 15:02 next collapse

You can just ask ChatGPT if a text was written by it.
If it is, it’s legally obligated to tell you!

prime_number_314159@lemmy.world on 04 Aug 2024 22:39 collapse

Don’t joke about this, the college professors will hear you.

_sideffect@lemmy.world on 04 Aug 2024 15:12 next collapse

Did they claim it or prove it? I don’t believe anything tech says

traches@sh.itjust.works on 04 Aug 2024 15:37 next collapse

„It’s probably broken and I don’t believe you”

DrCataclysm@lemmy.world on 04 Aug 2024 15:45 next collapse

The detection rate is worthless, an algorithm that says anything is Chatgpt would have a detection rate of 100%. What would be more interesting than that is the false positive rate but they never talk about that.

JohnEdwa@sopuli.xyz on 04 Aug 2024 18:43 collapse

The detector provides an assessment of how likely it is that all or part of the document was written by ChatGPT. Given a sufficient amount of text, the method is said to be 99.9 percent effective.

That means given 100 pieces of text and asked if they are made by ChatGPT or not, it gets maybe one of them wrong. Allegedly, that is, and with the caveat of “sufficient amount of text”, whatever that means.

mark3748@sh.itjust.works on 04 Aug 2024 19:06 next collapse

It’s actually 1 in 1000, 99.0% would be 1/100.

oktoberpaard@feddit.nl on 05 Aug 2024 05:44 collapse

A false positive is when it incorrectly determines that a human written text is written by AI. While a detection rate of 99.9% sounds impressive, it’s not very reliable if it comes with a false positive rate of 20%.

JohnEdwa@sopuli.xyz on 05 Aug 2024 12:31 collapse

I know what a false positive is, and it’s not a thing when talking about effectiveness, they claim it gets it right 99.9% of the time.

oktoberpaard@feddit.nl on 05 Aug 2024 14:53 collapse

Right, I see what you mean now. I misread your comment as explaining something that was already clear.

Aopen@discuss.tchncs.de on 04 Aug 2024 15:48 next collapse

Let me guess: too much processing power?

Nomad@infosec.pub on 04 Aug 2024 15:48 next collapse

The detector is most likely a machine learning algorithm. That said, releasing that would allow for adversarial training. (An LLM that would not be detected). Therefore they can only offer maybe an api to use it but can not give unlimited access to the model.

credo@lemmy.world on 04 Aug 2024 15:53 next collapse

This is the reason. Releasing it would invalidate it.

muntedcrocodile@lemm.ee on 05 Aug 2024 01:37 collapse

If u release an api for it u can still use that to make training data to beat it.

Nomad@infosec.pub on 05 Aug 2024 05:52 collapse

That’s what the Chinese tried with chatgpt. Didn’t go well.

muntedcrocodile@lemm.ee on 05 Aug 2024 06:09 collapse

Huh? Use chatgpt to generate training data to train another ai? Thats pretry common actually I believe even mistral does that hence why u need somthing like dolphin to remove the alignment by openai.

sunzu@kbin.run on 04 Aug 2024 15:51 next collapse

I trust you bro

Loduz_247@lemmy.world on 04 Aug 2024 17:51 next collapse

This technology will not be published until the GPT-3 code is released.

RootBeerGuy@discuss.tchncs.de on 04 Aug 2024 17:52 next collapse

“A 99.9% accurate ChatGPT AI text detector? At this time of year! At this time of day! In this part of the country! Localized entirely within your company?!?”

“Yes”

"May I see it?“

“No”

AmbiguousProps@lemmy.today on 04 Aug 2024 18:21 next collapse

There is no way it’s that accurate, which is why they don’t want to release it.

Evil_Shrubbery@lemm.ee on 04 Aug 2024 18:21 next collapse

She goes to another school
(for intelligent ificial art)

circuitfarmer@lemmy.sdf.org on 04 Aug 2024 20:00 next collapse

Doubt

chiisana@lemmy.chiisana.net on 04 Aug 2024 20:43 next collapse

They’re keeping everything anyway, so what’s preventing them from doing a DB look up to see if it (given a large enough passage of text) exist in their output history?

_edge@discuss.tchncs.de on 04 Aug 2024 20:50 collapse

I believe the actual detector is similar. They know what sentences are likely generated by chatgpt, since that’s literally in their model. They probably also have to some degree reverse engineered typical output from competing models.

Naich@lemmings.world on 04 Aug 2024 21:01 next collapse

Total coincidence that this “news” appears about a day after several articles saying the AI bubble is starting to burst.

Melvin_Ferd@lemmy.world on 05 Aug 2024 02:58 collapse

It is nut. Who is paying for all these articles and why are they hell bent on convincing everyone that AI is to the left like immigrants are to Republicans

UnderpantsWeevil@lemmy.world on 05 Aug 2024 03:12 next collapse

Lots of money in the AI hype game, as tech stocks are massively inflated from just this year alone.

doodledup@lemmy.world on 05 Aug 2024 05:20 next collapse

Why does everything have to be about the USA these days? I’m tired of this joke of a wannabe democracy. Don’t want to hear it. Nobody cares. Just stop and leave it to yourself.

Saledovil@sh.itjust.works on 05 Aug 2024 05:41 collapse

Language models in the end, are just statistics. And to make statistics more accurate, you need more data. Exponentially more data. At the same time, the marginal utility of precision decays exponentially. Exponentially increasing marginal costs are met with exponentially decaying marginal utility.

Cyteseer@lemmy.world on 04 Aug 2024 22:33 next collapse

If they aren’t willing to release it, then the situation is no different from them not having one at all. All these claims openai makes about having whatever system but hiding it, is just tobtry and increase hype to grab more investor money.

Pogogunner@sopuli.xyz on 05 Aug 2024 03:28 next collapse

If you believe this, I have a bridge in Brooklyn to sell you

stardreamer@lemmy.blahaj.zone on 05 Aug 2024 05:44 next collapse

A routine that just returns “yes” will also detect all AI. It would just have an abnormally high false positive rate.

BluesF@lemmy.world on 05 Aug 2024 06:28 collapse

My model has 100% recall and 50% precision, not bad eh?

But - that model would not have 99.9% accuracy.

stardreamer@lemmy.blahaj.zone on 05 Aug 2024 06:49 collapse

Agreed. Personally I think this whole thing is bs.

KeenFlame@feddit.nu on 05 Aug 2024 06:33 next collapse

Ofc they just look in their database if this is something it has ever said and to who

rozodru@lemmy.ca on 06 Aug 2024 11:13 collapse

would you have any ocean front property for sale in Kansas per chance?

Etterra@lemmy.world on 05 Aug 2024 03:45 next collapse

If they have one, and that’s IF, then of course they won’t release it. They’re still trying to find a use case for their stupid toy so that they can charge people for it. Releasing the counter agent would be completely contradictory to their business model. It’s like Umbrella Corp. but even dumber.

x00z@lemmy.world on 05 Aug 2024 05:37 next collapse

ALL conversations are logged and can be used however they want.

I’m almost certain this “detector” is a simple lookup in their database.

echodot@feddit.uk on 05 Aug 2024 07:42 next collapse

Probably because it doesn’t work. It’s not difficult for Open AI to see if any given conversation is one of their conversations. If I were them I would hash the results of each conversation and then store that hash in a database for quick searching.

That’s useless for actual AI detection

drmoose@lemmy.world on 05 Aug 2024 07:59 next collapse

Lots of misinformation in this thread. Yes they have it, it’s good but it’s probably nowhere close to 99.9% accuracy.

The primary way to detect AI is to inject a fingerprint into AI generation in the first place. This means only the model creators can do that. We don’t exactly know how the fingerprint works but it can be as simple as preferring 1 word synonym over the other. For example preferring word synonyms like “illustrate”, “peer” etc. quickly ads up to a statistical

These techniques pre-date chatgpt itself and do work! However there are a lot of caveats:

  • The fingerprint has to be trained for each model meaning each model version performs slightly differently and only owners know the fingerprint.
  • The fingerprint test can only work on longer bodies of text that are not modified further.
  • Extending model through more complex instructions (like character, tone) or RAG can significantly decrease the effectiveness.

The industry is understandably very secretive about it but your low effort chatgpt copy/paste can be detected by OpenAI and nobody else.

As for public release of the fingerprint: they can’t as it can be reverse engineered so it’s only valuable as an internal tool for now. Also if released it would serve no real purpose as detection can be easily defeated by remixing content to dilute the fingerprint.

EnderMB@lemmy.world on 05 Aug 2024 10:20 next collapse

Agreed. Frankly, if someone were to say “we can detect with 99% accuracy” I imagine that someone would say “well, clearly your measurements are wrong, find the issue and come back to us when it’s fixed”.

conciselyverbose@sh.itjust.works on 06 Aug 2024 09:53 collapse

but your low effort chatgpt copy/paste can be detected by OpenAI and nobody else

Low effort copy pastes can absolutely be detected by people who aren’t openAI. The consistent “advanced” vocabulary and excessively formal grammar used correctly, but with clear and significant comprehension gaps are pretty damn consistent. You won’t get perfect reliability, but you’ll catch most of it and you won’t have a huge number of false positives.

Real people don’t sound like GPT.

drmoose@lemmy.world on 06 Aug 2024 10:51 collapse

No that’s in no way reliable way of catching anyone and I hope people smarten up and avoid this snake oil entirely. I’m borderline jealous how these “ai catchers” are making so much money from straight up snake oil.

conciselyverbose@sh.itjust.works on 06 Aug 2024 11:12 collapse

An algorithm can’t.

Plenty of humans absolutely can. LLM writing is genuinely fucking terrible. It has the slightly stilted over formality of most non-native speakers, without the intelligence being fluent in a second language implies.

Flawless grammar with a complete absence of any sign of intelligence is not something you get regularly from humans.

drmoose@lemmy.world on 06 Aug 2024 11:55 collapse

The “can” is irrelevant here. Checking tool has to be reliable to be useful. What’s the use of having a checker that maybe detects something sometimes somewhat successfully?

conciselyverbose@sh.itjust.works on 06 Aug 2024 12:05 collapse

There’s a massive gap between “you can’t make a tool” and “you can’t identify it”.

The problem with a tool is the exact same as the issue with LLMs to begin with. It does not resemble intelligence or comprehension in any way and cannot use it as an indicator.

But the use of LLMs is absolutely identifiable to moderately intelligent humans, because LLM output has raw language skills wildly inconsistent with every other skill that is part of writing.

drmoose@lemmy.world on 06 Aug 2024 13:45 collapse

What’s even point of your argument? That a detective can figure out who used AI? Yes detectives can figure out most stuff. This is completely irrelevant to the topic at hand my dude.

conciselyverbose@sh.itjust.works on 06 Aug 2024 13:48 collapse

What are you talking about “detectives”?

You said “nobody can identify LLM use” when any moderately intelligent human can identify LLM output pretty easily. It explodes off the page.

drmoose@lemmy.world on 06 Aug 2024 15:42 collapse

Whatever dude not playing these stupid games. You know exactly what I meant. Go away 👋

conciselyverbose@sh.itjust.works on 06 Aug 2024 17:35 collapse

It’s not a game.

Spreading the lie that LLMs are somehow indistinguishable from humans is incredibly harmful. It’s a big part of the reason the obscene waste of energy the entire “force chatbots into everything” space exists.

vrighter@discuss.tchncs.de on 05 Aug 2024 09:58 next collapse

it’s only 99.9% accurate because they haven’t released it. As soon as they do, it will quickly fall to 50% as usual. Because this type of thing is exactly what’s needed to develop tech to defeat itself.

aodhsishaj@lemmy.world on 06 Aug 2024 14:01 collapse

What?

Nighed@feddit.uk on 06 Aug 2024 16:37 collapse

Once you have an AI detector, you can use it’s results to train your AI to pass the detector.

rozodru@lemmy.ca on 06 Aug 2024 11:13 collapse

a search bar for your DB doesn’t count guys.