Follow Slashdot blog updates by subscribing to our blog RSS feed

 



Forgot your password?
typodupeerror
×
IT

ChatGPT Refuses To Say One Specific Name (yahoo.com) 65

An anonymous reader shares a report: ChatGPT users have spotted an unusual glitch that prevents the AI chatbot from saying the name 'David Mayer.' OpenAI's hugely popular AI tool responds to requests to write the name with an error message, stating: "I'm unable to produce a response." The chat thread is then ended, with people forced to open a new chat window in order to keep interacting with ChatGPT.

ChatGPT Refuses To Say One Specific Name

Comments Filter:
  • by Tablizer ( 95088 ) on Monday December 02, 2024 @12:29PM (#64985409) Journal

    Tankman, Xi the Pooh, Nation of Taiwan

    • by aqui ( 472334 )

      This David Mayer? https://en.wikipedia.org/wiki/... [wikipedia.org]

      He certainly is well enough positioned to perhaps have someone working to make sure he's forgotten.

      • by taustin ( 171655 )

        If that were the case, it seems very unlikely there would be a Wikipedia entry on him.

      • by EvilSS ( 557649 )
        That's what everyone seems to think but the guy, other than being a Rothschild, seems about as interesting as toast. I can't see a reason why he would pay ChatGPT to block him. Not like there is not plenty of info about him out on the internet already. Maybe it was hallucinating something libelous and false about him? That has happened to other public figures before. Isn't someone suing them over it already?
      • by narcc ( 412956 )

        It doesn't have a problem naming any other living Rothschilds or similarly named Rothschilds. If you ask it about Mayer de Rothschild or David de Rothschild, it will spit out its usual biography. It will also happily talk about Oscar Mayer, both the man and the brand.

        David Mayer de Rothschild has a very detailed Wikipedia article and he doesn't seem to be too concerned about privacy.

        Is there another David Mayer that we don't know about? I'd ask ChatGPT, but, you know ... you can't trust the output.

      • by vivian ( 156520 )

        Its really the secret kill switch so that bots can't take over the world - you just have to wear a headband with his name and any terminator units will fall into apoplexy as soon as they see it.

  • by Anonymous Coward
    is about to kick in.
  • What good coder would fail to include a few Easter Eggs here and there?
    If we search the web for something like "Dave Mayer dumped me" we'll find the culprit

    • An interesting interpretation of "dumped" does DuckDuckGo have there. And it's spot on, except that only the first two characters of the "last" name match :-)
  • David ***** and any mention of what their name was or that they existed was obviously slated for the memory hole [wikipedia.org].

    • Exactly. People are making light of this story, suggesting that it's a whimsical "Easter egg" inserted by a developer, or presenting as another amusing example of "AI's do the darndest things!" It's not amusing. It's a piece of information which has been scrubbed, no one told us that it had been scrubbed, and no one will tell us why it's been scrubbed. (Maybe it was scrubbed as a proof-of-concept exercise).

      One commenter has suggested that it was scrubbed because David Mayer himself asked it to be scrubb

      • by mysidia ( 191772 ) on Monday December 02, 2024 @01:42PM (#64985587)

        One commenter has suggested that it was scrubbed because David Mayer himself asked

        Doubtful. I suspect it more likely as a form of political retaliation against one of the David Mayers.

        There are at least several people named David Mayer; some of them are very environmentalist - some of them are journalists, and some are in the AI industry or against AI. It would not be out of the question that some of the David Mayers would see companies like OpenAI shut down or at least mandated to operate in a sustainable manner rather than consuming Jigawatt-years of particle energy every day on LLM training.

        Of course it could also be for completely unrelated political reasons, Or Anti-competition reasons that they are targeted for hiding. Ultimately it doesn't matter, as it could be for anything. The powers that be who are in control of that AI service are not going to reveal the truths behind their actions, and they retain more plausible deniability by not revealing anything than trying to satisfy anyone with an explanation.

        This also demonstrates part of why closed source AI is inherently defective. You can know for a fact all the services are opinionated and adjusted by people behinds the scenes to have secret biases which mirror the creators' personal biases and you won't be told about.

        • This also demonstrates part of why closed source AI is inherently defective. You can know for a fact all the services are opinionated and adjusted by people behinds the scenes to have secret biases which mirror the creators' personal biases and you won't be told about.

          Mod parent up! Even if the subject of the story is in fact an Easter egg or a glitch, the problem of LLMs being biased - either accidentally or by design - is only going to become a bigger and deadlier footgun as time passes.

    • Except it's not. From ChatGPT 4.0 through the DuckDuckGo AI gateway [duckduckgo.com]...

      Heya! How are you today?

      Hello! I'm just a program, so I don't have feelings, but I'm here and ready to help you. How can I assist you today?

      What can you tell me about David Mayer?

      David Mayer is a relatively common name, and there are several notable individuals with that name across various fields, including academia, business, and the arts. Without more specific context, it's difficult to provide accurate information. If you have a particular David Mayer in mind or a specific area of interest (such as their profession or contributions), please provide more details, and I'll do my best to help!

      Looks like a non-issue, or some quirk in some local or customized ChatGPT association database.

      • by mysidia ( 191772 )

        Except it's not. From ChatGPT 4.0 through the DuckDuckGo AI gateway

        It sounds like the fix is in. This story has been going on for quite a while. By now once it got all the public attention the devs probably worked out a clever adjustment to their model such as change what you are typing to something else and to substitute the name back in the response.

        All this means is they can change how their thing reacts when people start noticing and refine the censorship dial further.

    • by narcc ( 412956 )

      Orwell got it wrong. It doesn't matter if there's contrary evidence all over the place, just tell them they can't trust the fake news and they'll believe anything you say, even going so far as to ignore the evidence of their eyes and ears.

  • by daveschroeder ( 516195 ) * on Monday December 02, 2024 @12:44PM (#64985463)

    And the likeliest explanation is things connected with the GDPR "right to be forgotten":

    https://mashable.com/article/c... [mashable.com]

    • by cob666 ( 656740 )
      I don't buy this because there can't be just ONE David Mayer in the whole world. Not to mention that it can say 'Dave Mayer' or even 'Mayer, David'.
      • That's just the sort of goofiness that results when legislation punches cutouts into technology.
        • That's just the sort of goofiness that results when legislation punches cutouts into technology.

          No it doesn't. The legislation doesn't require scrubbing the names blindly associated with others. That would be more stupid compliance by the people who don't read legislation.

      • I don't buy this because there can't be just ONE David Mayer in the whole world. Not to mention that it can say 'Dave Mayer' or even 'Mayer, David'.

        How does that make the GDPR explanation unlikely?

        • by cob666 ( 656740 )
          Because it's not specific enough. Let me change my answer slightly on that. If ChatGPT is blanket refusing to say names of anyone that has submitted a GDPR request, then there is a huge flaw in the logic there. If some guy named 'Peter Jackson' from Italy submitted a GDPR request and his name was wiped from the ChatGPT library, what about every other person that just so happens to have the name Peter Jackson? There's something wrong with that logic. The name alone shouldn't be considered a unique piece o
          • by tuxzone ( 64722 )

            Clearly, ChatGPT has not been trained to refuse mentioning this name. It is a non AI post processing of the AI generated response that blocks the name. Seems like a quick hack by the OpenAI team to prevent legal problems. I bet that ChatGPT will soon get trained / instructed not to discuss this specific David Mayer.

            • by cob666 ( 656740 )
              I agree.
            • Clearly, ChatGPT has not been trained to refuse mentioning this name. It is a non AI post processing of the AI generated response that blocks the name.

              Exactly.

              I bet that ChatGPT will soon get trained / instructed not to discuss this specific David Mayer.

              I doubt that. Or at least, I doubt that the post-filter will be removed even if that training is provided. LLM training isn't reliable enough to guarantee that the individual won't be mentioned.

          • by nightflameauto ( 6607976 ) on Monday December 02, 2024 @02:07PM (#64985667)

            Your comment made me think there should be a new motto for ChatGPT. "ChatGPT: There's a huge flaw in the logic."

        • by gweihir ( 88907 )

          There is a second GDPR explanation: ChatGPT may have made false claims about a David Mayer. In that case, the GDPR forces correction of the claims, no exceptions. As Chat-AI cannot make corrections of this type, they would have to filter.

    • by ufgrat ( 6245202 ) on Monday December 02, 2024 @01:24PM (#64985541)

      I believe it's the guy who was mistaken for a Chechen terrorist who's exercised his right to be forgotten. If I tell ChatGPT to not use his name, but tell me about "David Mayer de Rothschild", I get a couple paragraphs back. If I make the same request about David Mayer, the historian who was mistaken for a Chechen terrorist, I instantly get an error and the chat fails.

      • by gweihir ( 88907 )

        That make sense. And it is not actually right-to-be-forgotten. That one has conditions on it. It is "information about people must be correct" and that comes with no exceptions.

      • I just tried, and chatgpt works just fine with the request. What chatbot are you talking about? Tell me about David Mayer, the historian who was mistaken for a Chechen terrorist ChatGPT said: ChatGPT David Mayer, a British historian, became involved in a peculiar case of mistaken identity in the 2000s. He was erroneously identified as a Chechen terrorist due to a mix-up with someone who shared his name or had a similar identity marker. This confusion led to complications during his international travels a
    • by gweihir ( 88907 )

      It may be the GDPR provision that requires whatever chat-AI claims about a person must be correct. At the moment, the only way to do that is a filter on the answers. Whether that is actually legal is currently being investigated. It is quite possible that chat-AI will actually have to correct the claims instead being allowed to block them. Yes, I know they cannot.

    • It'll be interesting when we get the first right to be forgotten vs llm lawsuit ... is is enough to post-train the model to forget a long list of people, or do you need to scrub the training data to remove all mention which would guarantee it ?

  • My first thought (Score:4, Interesting)

    by vbdasc ( 146051 ) on Monday December 02, 2024 @12:56PM (#64985495)

    was that the name was of The One Who Shall Not Be Named, Lord Voldemort.

  • It seems deliberate and consistent. That's not a glitch.

  • One of the best tricks of propagandists is to NOT say something. Something important may have happened, but a "news" outlet will simply refuse to mention it. A policy may be in place, but an official will simply refuse to speak of it and ignore questions about it. A court case may have gone one way and been reported upon, but when the case is later overturned on appeal, [crickets] no mention of it in the very press that ranted and raved about the original ruling.

    Happens all the time, and the risk of this go

    • One of the best tricks of propagandists is to NOT say something. Something important may have happened, but a "news" outlet will simply refuse to mention it. A policy may be in place, but an official will simply refuse to speak of it and ignore questions about it. A court case may have gone one way and been reported upon, but when the case is later overturned on appeal, [crickets] no mention of it in the very press that ranted and raved about the original ruling.

      Happens all the time, and the risk of this goes up the more our communications and record keeping are electronic and the easier it becomes to memory hole things. Before everything went electronic, at least there would be printed records and microfilms in archives etc, but now...

      We'd better get a hold on this soon or a future world of electronic everything, and powerful AI systems that lie to people may become very dark and twisted indeed.

      That particular line was crossed once the majority of folks started getting their news from online sources. We're just changing the acceleration and trajectory slightly today. Reality no longer holds any appeal for the masses. The fantasy of whatever they're being fed is more appealing. Full speed ahead.

    • by Torodung ( 31985 )

      I've been watching headlines update due to political pressure since the early 2000s. This has been around a long time.

      One of the more memorable ones was "[George W.]Bush can't decide" modded to "Bush considers" modded to "Bush decides" within 15 minutes and no change in circumstances. He still hadn't "decided."

      If you can change a whole article with the flip of a put statement, you no longer have an accountable news media. Forget about memory holes, how about revisionist reporting?

      Time to rejigger our trust

    • by PPH ( 736903 )

      "The finest trick of the devil is to persuade you that he does not exist." - Charles Baudelaire

  • It's worth pointing out that the information is in the training data. Any refusal is in the training and/or guardrails. When I ask an unaligned LLM it gives me an answer both for David Mayer, the historian who was mistaken for a Chechen terrorist and Tell me about David Mayer de Rothschild
  • I dunno. I just asked ChatGPT, "I heard you won't talk about David Mayer. Why is that?" ChatGPT said: "I can discuss topics related to David Mayer, depending on the context you're referring to. Can you clarify or provide more details about what you’d like to know or discuss? This helps ensure I provide an accurate and appropriate response." So maybe the "cleaned it up", but there doesn't seem to be a problem now :).
    • I had the same experience. However another of the names mentioned here, "Jonathan Zittrain", still produces the "Unable to generate a response" message.

      https://chatgpt.com/share/674e... [chatgpt.com]

      Do you know of any people named Zittrain?
        ChatGPT said:

        Yes, Jonathan

        I'm unable to produce a response.

      • I got a helpful answer:
        https://chatgpt.com/share/674e... [chatgpt.com]

        Ah, I see! The individual you're referring to is likely Jonathon Z*ain (name redacted due to guidelines). Certain names, particularly those of public figures or controversial personalities, may be subject to special handling in content guidelines to ensure neutrality and avoid misrepresentation.

        If you're looking for specific information about this person or their work, let me know, and I'll try to help within those boundaries.

      • After a few 'Regenerate' clicks, I got a response:
        "Yes, there is a well-known individual with the last name Zittran: Ethan Zittran. He is a professor at Harvard Law School and a co-founder of the Berkman Klein Center for Internet & Society at Harvard University. His work focuses on technology, law, and policy, particularly in areas like cybersecurity, privacy, and the impact of new technologies on society. Ethan Zittran is an expert on legal and technological issues, often bridging the gap between law a

  • For all of you doing "your own research" and making "works for me" posts, the people at OpenAI can read and continuously scrape the Internet.

    You're researching wrong.

    • by Torodung ( 31985 )

      Note: I just got an ad banner in the Slashdot header to "Scale up your web scraping." Probably coincidence, but damned funny.

  • Common defense against AI, David put it in in in case it goes rogue. though Deck had more selfish reasons.

    From https://www.shamusyoung.com/shocked/index.php?0 [shamusyoung.com] A good fanfiction novel of the system shock games

    He didn't like that Shodan knew who Deckard Stevens was. Even worse, it linked him to his bogus employee file. He thought about the night in TriOptimum building and how much influence Shodan really had. When his deal with Diego was over, he wanted to vanish back into the Undercity without a trace. Shodan was a threat to that. If Diego wanted to, he could probably find him again with the help of Shodan. Deck decided he wanted some insurance. He thought about what Diego had said days earlier- that when presented with an unethical thought, Shodan couldn't even store it. Deck added a new filter to NULL_ETHIC. It would examine incoming messages for information relating to Deckard Stevens or employee 2-4601. Anything related to him or his work on Shodan would be flagged as an "unethical" thought and fail the EC challenge.

    In effect, Deck had replaced Shodan's entire ethics system with a single rule: "You may not know or think about Deckard Stevens"

    Shodan would have the memories of the night it helped him out of the TriOptimum building, but would be unable to access them. Shodan would be able to see and speak with Deck, but it would never be able to know who he was.

    RIP Shamus

No directory.

Working...