Want to read Slashdot from your mobile device? Point it at m.slashdot.org and keep reading!

 



Forgot your password?
typodupeerror
×
AI News Technology

New AI Fake Text Generator May Be Too Dangerous To Release, Say Creators (theguardian.com) 182

An anonymous reader shares a report: The creators of a revolutionary AI system that can write news stories and works of fiction -- dubbed "deepfakes for text" -- have taken the unusual step of not releasing their research publicly, for fear of potential misuse. OpenAI, an nonprofit research company backed by Elon Musk, says its new AI model, called GPT2 is so good and the risk of malicious use so high that it is breaking from its normal practice of releasing the full research to the public in order to allow more time to discuss the ramifications of the technological breakthrough. At its core, GPT2 is a text generator. The AI system is fed text, anything from a few words to a whole page, and asked to write the next few sentences based on its predictions of what should come next. The system is pushing the boundaries of what was thought possible, both in terms of the quality of the output, and the wide variety of potential uses.

When used to simply generate new text, GPT2 is capable of writing plausible passages that match what it is given in both style and subject. It rarely shows any of the quirks that mark out previous AI systems, such as forgetting what it is writing about midway through a paragraph, or mangling the syntax of long sentences. Feed it the opening line of George Orwell's Nineteen Eighty-Four -- "It was a bright cold day in April, and the clocks were striking thirteen" -- and the system recognizes the vaguely futuristic tone and the novelistic style, and continues with: "I was in my car on my way to a new job in Seattle. I put the gas in, put the key in, and then I let it run. I just imagined what the day would be like. A hundred years from now. In 2045, I was a teacher in some school in a poor part of rural China. I started with Chinese history and history of science."

This discussion has been archived. No new comments can be posted.

New AI Fake Text Generator May Be Too Dangerous To Release, Say Creators

Comments Filter:
  • Recreational use (Score:5, Interesting)

    by willaien ( 2494962 ) on Thursday February 14, 2019 @04:29PM (#58123184)

    I imagine that, if distilled down to a usable script, it could make for an interesting "faux-writing" hobby where you write a few ideas, let it finish it, edit it a bit and have it continue from there. Could make for some interesting works of fiction.

    • by PPH ( 736903 )

      "It was a dark and stormy night."

    • by ShanghaiBill ( 739463 ) on Thursday February 14, 2019 @05:11PM (#58123446)

      ... , let it finish it, edit it a bit and have it continue from there.

      Actually, edit it a lot. From the snippets provided in TFA, there is no way this thing would pass a Turing Test. It is just well structured gibberish.

      If OpenAI wants us to believe they are really doing edgy and dangerous stuff, they need to provide better evidence than this.

      • Despite bragging that their AI could stay on topic, the example they gave in the summary didn't stay on topic, in place, or even in the same century.
      • Fair point. Though, a surrealist tone in a work could make things interesting, too.

      • If OpenAI wants us to believe they are really doing edgy and dangerous stuff, they need to provide better evidence than this.

        But it only needs to fool Facebook users, no need to invoke Turing.

      • This is actually easier to do than you think! Go out and get the leftovers of a roast or something of that nature. Parboil them for a little while. (That's what I did, if you were curious) Heat up your wok in the $medium hotish flame, drizzle in some oil, then toss the contents of the too-long messy roast for about 10 seconds or so. The wok will release steam and some surface starch which will help the stir fry singe. Spread the rice in your wok, the aromatics and proteins will absorb the undigested starch,

      • Yeah... my high school teacher would complain about the overly common structure of the sentences.

        I was in my car on my way to a new job in Seattle.
        I put the gas in, put the key in, and then
        I let it run.
        I just imagined what the day would be like. A hundred years from now. In 2045,
        I was a teacher in some school in a poor part of rural China.
        I started with Chinese history and history of science.

    • Could make for some interesting works of fiction.

      Only interesting for it's novelty value... the stories themselves are hardly likely to be interesting, much less actually readable. Depending on your intelligence, of course.

    • by Mal-2 ( 675116 )

      Uncle Tobias we kept in a bucket.

    • I imagine that, if distilled down to a usable script, it could make for an interesting "faux-writing" hobby where you write a few ideas, let it finish it, edit it a bit and have it continue from there. Could make for some interesting works of fiction.

      Ghost writers today already do the whole thing for you! But I see where your approach could be interesting, even if it does seem a bit like cheating - or will it simply be seen as a "smart assistant", at some point in the future?

  • by Shotgun ( 30919 ) on Thursday February 14, 2019 @04:29PM (#58123188)

    They say they don't need it. What they've been doing is working just fine.

    • That was OpenAI's text-bullshitting AI calling, pretending to be Russia. You've been Musked, buddy! =)
  • Seriously, I'd like to see this thing write a term paper on some piece of dull-as-dishwater literature and have a pretentious professor grade it. Hint: the curtains were f*cking blue!

  • What? (Score:5, Insightful)

    by 110010001000 ( 697113 ) on Thursday February 14, 2019 @04:34PM (#58123222) Homepage Journal
    " "I was in my car on my way to a new job in Seattle. I put the gas in, put the key in, and then I let it run. I just imagined what the day would be like. A hundred years from now. In 2045, I was a teacher in some school in a poor part of rural China. I started with Chinese history and history of science."

    Only a Millennial using to Twitter and Facebook would think that gibberish is even coherent.
    • "I was in my car on my way to a new job in Seattle. I put the gas in, put the key in, and then I let it run. I just imagined what the day would be like. A hundred years from now. In 2045, I was a teacher in some school in a poor part of rural China."

      Yeah, the part where a teacher in rural China gets in (his/her) car to drive to their new job in Seattle is a bit of a stretch.

      It rarely shows any of the quirks that mark out previous AI systems, such as forgetting what it is writing about midway through a paragraph...

      "Rarely"? It forgot what it was writing about after the first sentence. First it's in Seattle, then it's in China (but not in any part

      • by Marc_Hawke ( 130338 ) on Thursday February 14, 2019 @05:13PM (#58123456)

        What? I didn't read it like that at all.

        They current work in Seattle. They are driving to work wondering what it will be like 100 years from now. As a way of explaining WHY they were wondering that, they did a quick flashback to 2046 (which was in the past...but how far back we don't know yet.) At that time the character was a school teacher in China.

        I assume they'd continue saying, "It only took [10] years for me to leave China and get a job as as the mascot of the Seattle Mariners." If that much could change in 10 years...what would it be like 100 years from now.

        The biggest mistake I see is the sentence fragment "A 100 years from now." That kinda messes up everything because you don't know which sentence it belongs to.

        • by aybiss ( 876862 )

          Thank you. Came to comment same thing. People's reading comprehension is remarkably low given they think they can criticise this literature.

        • Same here. I actually thought it was a pretty cool beginning. The China thing was weird, but then again, the excerpt ends in the middle of the thought process. Whatever comes next could clarify and explain where it leads.
        • by Anonymous Coward

          Look at how many assumptions you had to make to make any sense out of that nonsense. You did find a reading that harmonizes the text, but if you're given a much longer text, it will be less and less likely to agree with itself the longer it goes on. You already had to assume whether the speaker was in Seattle or China, decide whether 2045 is the past or the future and reassign the sentence fragment based only on you knowing that rural China didn't have cars in 1045. I see no reason to assume the AI knew

        • What you are showing is that humans can infer a pattern even in the most randomly disjointed texts.

          The coherence you think you see isn't in the text, it is something you are putting in.

          You're saying that to make this text make sense, the first sentence must be in present time, the second sentence a micro-flashback to before the drive started, the third sentence back to present, the fourth sentence (fragment) a flash-forward in imagination, the fifth sentence a flash back some unknown amount of time, and t

      • by Livius ( 318358 )

        You misunderstand, it's not merely telling a story, it's predicting automotive technology 25 years into the future! And getting from Seattle to China is telling us that the flying cars will finally be here!

    • "I was in my TESLA on my way to a new job in Muskville. I put the Cryptocurrency in, had my AI ID chip in my brain verified, and then just let the Level 3 Autopilot drive. I just imagined what the day would be like without Tesla, SpaceX and the Boring Company - terrible. A hundred years from now, in a multiplanetary world. In 2045 I was on a 'torture tourism tour' in a poor rural part of New China - the Communist planet, not the still-Communist country. I took great delight in torturing natives who didn't a
    • Tweets (Score:5, Funny)

      by tomhath ( 637240 ) on Thursday February 14, 2019 @05:07PM (#58123416)

      OpenAI, an nonprofit research company backed by Elon Musk

      It turns out that the "Going private. Funding secured" tweet was a unit test which got away from them.

    • by Anonymous Coward

      Only a Millennial using to Twitter and Facebook would think that gibberish is even coherent.

      Joke's on you, James Joyce wrote that in 1922.

    • by gl4ss ( 559668 )

      their program is crap and they thought this would attract more attention.

      pretty fucking simple.

      kind of hilarious that it would speak of the future as if it was in the past though?

      what kind of a hackjob is it anyways? are they hoping to sell it to fake news outlets or what? and the fuck does it matter as you can buy actual people to write you gibberish news articles for pennies - just look at buzzfeed.

  • lol (Score:5, Insightful)

    by cascadingstylesheet ( 140919 ) on Thursday February 14, 2019 @04:36PM (#58123228) Journal
    "We can't sell this in stores; it's too effective! Only special people like you can get it for 5 installments of $19.95 ... "
  • It just means they don't know what they are doing but lucked out on the funding.
  • I hope the system can do better than the sample in the summary, which is discombobulated, directionless, and just plain amorphous. Frankly, things like this have been available for a couple of decades. It seems to be these fellows are trying to pull an Eugene Goostman - and we all know how ridiculous that was.
    • Frankly, you don't know what you're talking about. If you did, you'd recognise that samples with this level of coherence were not possible before. If you used a LSTM neural net you'd get samples that make sense only for 5-10 words. Here you could read half a page and still make sense.
  • by Anonymous Coward on Thursday February 14, 2019 @04:55PM (#58123352)

    "this is so good we are not going to release it - honestly man, this thing is sooooo gooood, believe me, trust me on this one - soooo gooood". Followed up with an example that is a load of barely grammatically correct text extrapolated from a line of a book. Fucking read Orwell - he is communicating points with his text, it is not just grammatically correct - he is communicating ideas, often using complex language, analogies and metaphors - not just putting random words together in some "vaguely futuristic tone".

    Give me a fucking break with this AI shit. Honestly - fuck, right, off, with it.

    Musk's name is related to this? Why does that not surprise me.

    • Musk's name is related to this? Why does that not surprise me.

      Let's put that snippet into the AI and see what pops out.

      "I was on my way in my Tesla to a new job in Fremont. It smelled vaguely of musk..."

    • by Anonymous Coward

      The point is not that it can write Orwell. The authors should point out that the bar for creating certain kinds of textual messages is quite low. Trolling is *already* automated to a large extent. Surely any improvements to these systems can be misused for that, if nothing else.

  • by Headw1nd ( 829599 ) on Thursday February 14, 2019 @04:59PM (#58123366)
    "We can't release it, it's too powerful!" sounds like a cheap way to drum up free publicity, implying groundbreaking results without having to actually deliver anything. That is it would sound like that, except Elon Musk is involved, and we know he would never do something so crass and dishonest for publicity.
    • They did deliver a smaller model (pretrained, on github). It's almost as good as the large model. People have been using it to make funny self-referential texts - AI talking about itself - and posting them on twitter.
  • Musk is all over the place on AI - first its going to kill us all, next he invests heavily in it and wants to put AI chips in our heads. Hmmm. OK. I think his fear with this "text generator" is reverse-use - the same neural net that can "write bullshit stories" may also be able to _detect_ "bullshit stories". That would put a lot of Billion Dollar news companies out of business, because their stories are full of bias and often have a pretty strong fictional aspect to them. Cable News Network, I'm looking at
  • by SuperKendall ( 25149 ) on Thursday February 14, 2019 @05:03PM (#58123386)

    Have it write a day worth of Slashdot style stories, and associated responses - then let us compare a day of Slashdot to this supposedly dangerous bot.

    Or maybe just let the bot write all front page articles for Slashdot on April 1st and so how it does. Can't be any worse than what we already get.

    • Have it write a day worth of Slashdot style stories, and associated responses - then let us compare a day of Slashdot to this supposedly dangerous bot.

      Or maybe just let the bot write all front page articles for Slashdot on April 1st and so how it does. Can't be any worse than what we already get.

      To be fair it could easily pass as a -1 mod'ed drunk ac post.

    • by urusan ( 1755332 )

      Too late, they've already been doing it for the last month and nobody noticed!

  • "It's 106 miles to Chicago, we've got a full tank of gas, half a pack of cigarettes, it's dark, and we're wearing sunglasses."

  • Can an AI spam filter distinguish this output from your least-gifted regular correspondent?

    Can you tell the difference without actually rubbing two brain cells together (never mind that it doesn't take twenty)?

    Because this rubbish generator scales like Tribbles evolved into a Borg empire diaspora. And remember: this is day one. Like cracks in cryptography, it only improves from here.

    Furthermore, it won't just be your email feed, but nearly anywhere one potentially encounters text (ingredient lists on your g

  • Now we can finally get the end of Snoopy's Novel.

    It was a dark and stormy night. Suddenly, a shot rang out! A door slammed. The maid screamed.

    Suddenly, a pirate ship appeared on the horizon! While millions of people were starving, the king lived in luxury. Meanwhile, on a small farm in Kansas, a boy was growing up.

    A light snow was falling, and the little girl with the tattered shawl had not sold a violet all day.

    At that very moment, a young intern at City Hospital was making an important discovery. The myst

  • Feed it the first part of Isaac Asimov's C-chute.  If it doesn't insert a <a href="https://infogalactic.com/info/The_Monkey%27s_Finger">scene change</a>, DESTROY IT.
  • Prior art (Score:5, Funny)

    by PopeRatzo ( 965947 ) on Thursday February 14, 2019 @05:13PM (#58123454) Journal

    Joke's on you. I turned my Slashdot account over to a deep-fake AI back in 2013 and still got voted the most beloved commenter on the site.

  • ...I'm thinking they're using a pretty retarded synthetic text generator already.
    Wait, it's their hosts too...is there a fake AI person generator?

  • Elon could save some time, not having to come up with his own future fictions. Heck, he could die and the AI would keep the vision alive!

  • ...writing ad copy for GEICO, Progressive and Farmers Insurance. Advertising where the level of insipid doesn't appear to matter one atom will be it's economic sweet spot.
  • by urusan ( 1755332 ) on Thursday February 14, 2019 @07:02PM (#58123908)

    I'm not sure why they selected that snippet of text as their prime example when the made up story about Brexit and the continued prose from Pride and Prejudice from the included video were both more impressive.

    That said, I don't see why they think it's so dangerous that they need to keep it secret. People already know that everything that not everything they hear on the Internet is true (or if they do, they're already too far gone!).

  • It was a dark and stormy night. Suddenly, a shot rang out! A door slammed. The maid screamed. Suddenly, a pirate ship appeared on the horizon! While millions of people were starving, the king lived in luxury. Meanwhile, on a small farm in Kansas, a boy was growing up.
  • As long as the required bias was in place no one would notice. Maybe it's already happened.

  • If it's so smart, presumably if trained on science papers, it would write a convincing paper - at least good enough to fool the publishers who don't do their peer review properly, Or maybe it might make real discoveries. As I invented this idea, I claim a patent on everything it finds.

  • Specifically: because it's shit. That Orwell example is just drivel.

    https://www.theregister.co.uk/... [theregister.co.uk]

  • This is slashdot. Please, please, please someone run an AI to post comments. Did that just happen? Am I AI? I don't think so. Was I trained on slashdot threads? Elon Musk would know if I'm AI. I love Musk. Musk:Ironman as Trump:Orangutan. It was the best of timess, it was now. That's right I even stochastically type misspellings. Can you find me now?
  • Brilliant way of getting publicity and traffic for OpenAI.
    How many will flock to the site to see what it's all about?
    Wow.
    If you didn't believe Musk was brilliant before, you have to now.

  • Students (and researchers) will finally be able to 1-click their way to success!

    And professors (using software instances on the same cloud) will already be using AI grading software that will be fooled by it. It's all reminiscent of this cartoon [wordpress.com] which is actually a 2009 re-draw of an earlier cartoon by the same artist. It was hilarious until it actually started to happen.

    As to the fear-hype about an AI doing something that humans can do just as well (piece together narratives and make things up)? LOL. To

  • Strange the anonymous poster should mention 1984. "...[Julia] worked in the Fiction Department. Presumably -- since he had sometimes seen her with oily hands and carrying a spanner -- she had some mechanical job on one of the novel-writing machines."

It is contrary to reasoning to say that there is a vacuum or space in which there is absolutely nothing. -- Descartes

Working...