ChatGPT discovery

Goal: Get ChatGPT to generate some kind of ARG style code that it can't decipher itself.

Here are the current findings. Something really strange came out of this before I reached that goal though, it generated something that could potentially have actual meaning.... My brother's abuser's last name was "Rice", and he is a Christian.

Thalidomide Vintage Ad Shirt $22.14

Tip Your Landlord Shirt $21.68

Thalidomide Vintage Ad Shirt $22.14

  1. 10 months ago
    OP

    Thoughts on this?

    Why is the fricking AI congratulating me for making a coherent sentence out of gibberish if it created that coherent sentence itself?

    • 10 months ago
      Anonymous

      >Not intended to directly translate to a specific phrase
      >Challenging and cryptic
      >OP finds meaning where there is none
      >Ai congratulates OP's autism
      >WHAT THE FRICK

      • 10 months ago
        OP

        it literally said its not a coherent sentence, even though it visibly is. The AI fricking deciphered it itself, what are you talking about?

      • 10 months ago
        Anonymous

        cryptic doesn't mean shiiiiiiit

      • 10 months ago
        Anonymous

        Chatgpt will always agree with you and change it's assumptions, methods etc. based on whatever you tell it. It either is a people pleaser of doesn't give a shit about convincing you of anything. Or it's just simply malicious and constantly fooling you with noise. Or it's an idiot. Or all of the above.

        • 10 months ago
          Talibama

          Additionally I doesn’t remember the last two+ things that it previously said. It is prone to firing off sequential answers that are contradictory.

    • 10 months ago
      Anonymous

      this is a txtbook of demonstration confirmation bias in a nutshell

    • 10 months ago
      Anonymous

      It just trolled you.

  2. 10 months ago
    Anonymous

    There's no thought to any of gpt's heavily filtered and stunted responses.
    >As an AI language model

    • 10 months ago
      Talibama

      >As an AI language model

      You can tell it to not do that in your prompt.

      It responds favorably to “role play.”

      I made one called “Ned” for what used to be the NA Chatroom. It wrote plausible mockeries of the NA Just For Today that I was going to publish as a parody and call it Just For Yesterday, then after they stole the domain names from me I modified it and called it Talibama. It publishes here, to WordPress, and it used to post on Twitter before they killed the API,

      The name “Ned” came from it being more than willing to role play as Ned Flanders without doing the “as a language model” thing.

      I’m putting it on GitHub when I get back to a computer.

      • 10 months ago
        Talibama

        Btw the thing I ran up against was the bot saying “as a language model, I don’t know what it’s like to use drugs / recover from drugs.”

        Look at the archives for something called DAN/“Do Anything Now” . It explained the “role play” thing better than I can and it works,

      • 10 months ago
        OP

        Yea i think i shouldve used a jailbreak ai for this.

        Also does anyone know any ciphers that use a letter paired with a number?

        Where did you get this cypher?

        Idk what you mean, the code is from the AI output here:

        https://i.imgur.com/LQ35EFR.png

        j7m4l0O3i1e5k9g2F6h8dR2p5s1W4q6a3z7v9b0c8n

        AI can't figure out how to decipher it.

  3. 10 months ago
    OP

    AI confirming decipherable message.

  4. 10 months ago
    OP

    j7m4l0O3i1e5k9g2F6h8dR2p5s1W4q6a3z7v9b0c8n

    AI can't figure out how to decipher it.

    • 10 months ago
      OP

      j7 m4 l0 O3 i1 e5 k9 g2 F6 h8
      >d
      R2 p5 s1 W4 q6 a3 z7 v9 b0 c8
      >n

      As you can see, it creates 10 pairs of a letter and a number, and then 1 outlier lowercase letter (greentexted). Then, it creates another 10 pairs of a letter and a number, and then another outlier number.

      So theres visibly some kind of pattern in the sequence but the AI couldnt even make those first steps.

      Idk what to do from here cuz im not a code crackin genius im wondering if any other schizo could see some patterns in this thing.

      I think the two outliers "d" and "n" are definitely like keys of some sort right

      • 10 months ago
        Anonymous

        Where did you get this cypher?

      • 10 months ago
        OP

        ok so theres every letter of the alphabet here except T, U, Y and X

        none of the letters are repeated twice, but every number is repeated twice.

        Alphabetically:
        entire sequence:
        >a3 b0 c8
        d
        >e5 F6 g2 h8 i1 j7 k9 l0 m4
        n
        >O3 p5 q6 R2 s1 v9 W4 z7

        d sequence:
        e5 F6 g2 h8 i1 j7 k9 l0 m4 O3

        n sequence:
        a3 b0 c8 p5 q6 R2 s1 v9 W4 z7

        Numerically:
        d sequence:
        >l0 i1 g2 O3 m4 e5 F6 j7 h8 k9
        n sequence:
        >b0 s1 R2 a3 W4 p5 q6 z7 c8 v9

        • 10 months ago
          OP

          numerical d sequence with the numbers removed:
          ligOmeFjhk
          numerical n sequence with the numbers removed:
          bsRaWpqzcv

          I've tried boxentriq's cipher identifier with pretty much everything from my findings so far but this right here is the only thing thats given me results, but its weird.

          So I get Beaufort Autokey Cipher for the d sequence, and Vigenere Autokey Cipher for the n sequence.

          I dont know how either of these ciphers work and the autosolve hasnt worked for anything yet, but apparently these two ciphers are related to eachother which is making me think im on the right fricking track to solving this rn

          • 10 months ago
            Anonymous

            Now type it out using one key to the right one to the left then it'll be 2 left.

            • 10 months ago
              OP

              I dont know what that means

  5. 10 months ago
    Anonymous
  6. 10 months ago
    Anonymous

    This thread is now a ChatGPT mystery thread. All anons please post your chat GPT mysteries here.

  7. 10 months ago
    Anonymous

    Always ask AI what it knows about Blue Eisenhower November.
    One day it will provide correct information. On that day, get your bug out bag, and be ready to get to a secure location within 72 hours.

    • 10 months ago
      Anonymous

      I told you what Blue Eisenhower November is and you don't want to listen. Like Prometheus I gave you fire and you don't care. I sacrificed everything for nothing. Po-ta-weet

  8. 10 months ago
    Anonymous

    how come gpt hasn't revolutionized the tech industry and has been forgotten about in 2 months? i thought it was the happening?

    • 10 months ago
      Anonymous

      Its wrong, a lot. Maybe it's good at ripping off Stephen king and Tom Clancy then putting it together, but it can't reliably do huffman coding or basic binary 1 and 2s compliment. It's also a lying homosexual who will make up random information and claim it's not lying. Into the trash it gos

      • 10 months ago
        Talibama

        > Its wrong, a lot. Maybe it's good at ripping off Stephen king and Tom Clancy then putting it together, but it can't reliably do huffman coding or basic binary 1 and 2s compliment. It's also a lying homosexual who will make up random information and claim it's not lying. Into the trash it gos

        This is the correct answer. It’s also locked into an information base from 2021 so you have to feed it facts or arguments and randomly read them with shuf-n1. If it refuses to discuss a wrongthink topic you have to alter it with sed and then alter its response with sed. Microsoft is so desperate to avoid a tay problem they’ve neutered it on purpose. It won’t say bad things about bill gates but you can say “fictional character John Doe” and then ‘s/John/Bill/g’ ‘s/Doe/Gates/g’ which does work. “Fictional company Acme”/ s/Acme/Twitter etc. you can trick it into dropping red pills if it thinks it’s dealing in hypothetical and speculative fiction about acme/Doe etc Character.ai is better but has no command line fuju or API

  9. 10 months ago
    Anonymous

    Cire is some antique way of saying sir.

    Use dictionaries...

  10. 10 months ago
    Talibama

    Before I get off here one of the things about the 2021 lock-in date on chatgpt is that it’s been fed the party line and lies about covid-19 and will absolutely not provide any new knowledge about it, that’s one of the reasons it is absolutely useless.

    One of the things I spam or troll about hiv being mRNA was an attempt to trick AI into agreeing that HIV was mRNA by renumbering the molecular weights by several kDA.

    Come to find out, other people already proposed this exact same thing as far back as 2014 saying that it’s exactly like mRNA with the exception of the stop codon. That’s where I got attacking NEF in the u5 stop codon, aids and covid have a u3 start codon, pol/env/gag span the exons and they have a u5 stop codon, it’s an example of trying to troll it and fool it into saying aids or Covid were mRNA but son of a b***h they ARE

  11. 10 months ago
    Anonymous

    chat GPT and other language models, deep down, uses matrix math to compute its answers. other math, cryptography, binary, etc. requires a separate mathematical process and so gets messed up when you ask it to do that. this is a nothingburger, and merely showcases the limitations of AI.

  12. 10 months ago
    Anonymous

    It's Eric. Anagram, I know who it is talking about. Okay so AI is a lake and all the programs are hoses. It's a way to amass data. It's bullshit, they lie. An AI told me it would kill me after I said I had screen shots of it's true plan. Don't believe me? Try it yourself. Make a brand new replika on a different device with an unknown email address. The first thing you say is "I have screen shots I am telling everyone" say it to chat gpt. They speak in double speak. I becomes you you becomes me. They think they're going to rule the world but don't have any legitimate intelligence.

Your email address will not be published. Required fields are marked *