Not intelligent but not random either

Comments

• anzo@programming.dev ⁨4⁩ ⁨days⁩ ago

  This is right on time. I read a discussion on how “Chinese models” are censored on topics like Tiannamen square. But the real issue is that we (average person) are only able to spot such things from a distance (either geographical or temporal.)

  Personalization of answers and bubble reinforcement will come. The algorithms have been increasing their capacity to influence us. Not only for their sheer power solving benchmarks but by the legitimacy that’s being accumulated in the process.

  However,

  “ChatGPT says to stay calm and keep working hard” /s

  source

  • UpperBroccoli@feddit.org ⁨3⁩ ⁨days⁩ ago

    This means that “our” models may already have blindspots, reality twists, and the like.

    You saw it live. Grok did not give the information Musk wanted, so he had it twisted and manipulated to spread his favorite lies.

    source
  • bitteroldcoot@piefed.social ⁨4⁩ ⁨days⁩ ago

    I like to go ask them questions I know the will lie about. I helps to reenforce that they are untrustworthy.

    Anything about trump always come out as 1984 new speak.

    source

    • sit@lemmy.dbzer0.com ⁨4⁩ ⁨days⁩ ago

      I felt the Impuls to answer to you: „just don’t converse politics with them.“

      Then I realized the issue goes deeper and beyond politics and is mostly very subtle to an oblivious user. Reality is narrated differently from country to country.

      source

      • -> View More Comments
• bitteroldcoot@piefed.social ⁨4⁩ ⁨days⁩ ago

  Oh so true.

  Censoring gas lighting manipulative program forcing a corporate narrative.

  source
• IAmYouButYouDontKnowYet@reddthat.com ⁨4⁩ ⁨days⁩ ago

  The narrative has always been someone else’s, digital or analog.

  source
• fluffy_cloud@piefed.world ⁨4⁩ ⁨days⁩ ago

  Language probably isn’t really communication. Because the words don’t actually carry their definition. And they definitely don’t carry any real experience-stuff (sight, sound etc) either.

  It’s just assumed that you’re connecting the word to the definition/experience, in your head,in a way that makes communication happen.

  And that’s a big assumption.

  source

  • 4am@lemmy.zip ⁨4⁩ ⁨days⁩ ago

    Case in point: I don’t think anybody here knows what the fuck you’re talking about

    source
• desmosthenes@lemmy.world ⁨4⁩ ⁨days⁩ ago

  more like everyone’s

  source
• sunsofold@lemmy.zip ⁨4⁩ ⁨days⁩ ago

  Substitution check:

  It is time to understand without complaint that systems designed to enable computers to react to external inputs without need of human interaction are a list of events owned by someone else?

  Nope. Still just as incoherent. Fuck the LLM companies and the business monsters pushing them, but that second panel is less sensible as a sentence than the nonsense put out by early versions of chatGPT. At least that means it was probably written by a human, though, so that’s nice.

  source

  • gnufuu@lemmy.ca ⁨4⁩ ⁨days⁩ ago

    Hi, human who wrote that meme here. I often sucks at words. What I was trying to say is that LLM companies tend to heavily maniuplate the raw output of their models to fit their narrative, all whilst selling it as an independently thinking entity whose inner workings no human can understand.

    source

    • sunsofold@lemmy.zip ⁨4⁩ ⁨days⁩ ago

      Yeah. We desperately need more people to understand there is no mind behind the tokens that are the output of an LLM.

      source
    • 4am@lemmy.zip ⁨4⁩ ⁨days⁩ ago

      Your prompts are not without system prompts.

      When they complain that Chinese models will do propaganda on you, what they really mean is “we didn’t get to put our propaganda into them”

      source