Login
You're viewing the mastodonapp.uk public feed.
  • Darius Kazemidarius@friend.camp
    Oct 25, 2026, 6:56 PM

    One thing I learned from making Twitter bots and web toys with "personality" for years is that some people *desperately* wanted those things to be intelligent. I would say exactly how my algorithm worked to show that it was doing nothing more than pulling random items from a set. But some would assume there was magic happening on some level anyway. There was a hunger for that to be the case, and I think LLM companies trade in sating that hunger.

    💬 12🔄 63⭐ 0

Replies

  • Darius Kazemidarius@friend.camp
    Oct 25, 2026, 7:01 PM

    For example, "Content, Forever" generates shitty essays on a given topic prompt by following links on Wikipedia and pasting whole paragraphs from each article into the output.

    tinysubversions.com/contentFor

    I had so many inbound requests from high schoolers asking if I could make it do their homework. I would sometimes write back and explain, no, this isn't how this works, it's explained in the heading if you want to know more.

    Now they finally have the homework machine, I guess.

    💬 1🔄 3⭐ 0
  • Julien Deswaefjulien@xuv.be
    Oct 25, 2026, 11:13 PM

    @darius Totally. As Weizenbaum discovered himself: "I had not realized ... that extremely short exposures to a relatively simple computer program could induce powerful delusional thinking in quite normal people."

    💬 1🔄 3⭐ 0
  • Julien Deswaefjulien@xuv.be
    Oct 25, 2026, 11:18 PM

    @darius It's actually the randomness that makes these LLM feel like magic, that "temperature" factor as they call it. Turn it to 1, remove randomness, and they appear pretty stupid. So I'm not surprised that random bots have totally felt "intelligent" to people. People back then also asked if Shiv Integer was AI.

    💬 0🔄 1⭐ 0
  • Benjamin Geerbenjamingeer@piaille.fr
    Oct 25, 2026, 7:09 PM

    @darius In _The Way We Think_, Fauconnier and Turner argue that human conceptualisation strives to reduce abstraction and complexity to "human scale", and anthropomorphism is one very common way to do that.

    There is one overarching goal driving all of the principles: * Achieve Human Scale. The constitutive and governing principles have the effect of creating blended spaces at human scale. The most obvious human-scale situations have direct perception and action in familiar frames that are easily apprehended by human beings: An object falls, someone lifts an object, two people converse, one person goes somewhere. They typically have very few participants, direct intentionality, and immediate bodily effect and are immediately apprehended as coherent.
    💬 0🔄 1⭐ 0
  • Random Geekrandomgeek@masto.hackers.town
    Oct 25, 2026, 7:12 PM

    @darius and worse yet they get high on their own supply and the whole thing ends up like dogs telling knock knock jokes

    "knock knock—"

    "BARK BARK BARK"

    "BARK BARK BARK"

    (apparently a sufficiently old joke that I found five old comics with the exact template so I'll just throw this one from They Can Talk instead)

    First panel: dog resting, hears a knock at the door Second panel: dog goes to the door and asks "who's there?" Third Panel: dog is very close to the door. "I said who's there?" Final panel: dog is angry now: "DON'T YOU KNOW HOW THESE JOKES WORK?"
    💬 0🔄 1⭐ 0
  • 💬 1🔄 0⭐ 0
  • 💬 1🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • Joltrast :mastodon: :clippy:joltrast@mastodon.social
    Oct 25, 2026, 8:29 PM

    @darius I used to configure IRC Perl chatbots back in the day. Not too dissimilar to ELIZA. I still have the manual knocking around somewhere. I think it really helped me understand "how the sausage was made" even back then. Classic "If This Then That". Not terribly sophisticated, but efficient.

    💬 1🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • James Tweedie :yes_scotland: :noai:MyricaGale@mastodon.scot
    Oct 25, 2026, 8:55 PM

    @darius Unfortunately it's been like that since Eliza in the late '70s.
    I remember "introducing" non-computer friends to Eliza and them being amazed by the "magic" of a system that responded to their questions! Current developments are massively more "advanced".

    💬 0🔄 0⭐ 0
  • webhat🔜#40c3webhat@infosec.exchange
    Oct 25, 2026, 9:00 PM

    @darius I once wrote an RSS feed reader with the most simple Bayesian algo that took keywords from the title and article and recommended new articles based on previous hits

    It was quite accurate in recommending new articles I hadn't read that might interest me, if I hadn't have written it myself I would have thought there was a more complicated algorithm behind it

    💬 0🔄 5⭐ 0
  • 💬 0🔄 0⭐ 0
  • Bradley Mombergerair_hadoken@mastodon.social
    Oct 26, 2026, 3:55 AM

    @darius I got so many people arguing with my bot that just reposted issue titles from Github, that I added an auto-reply that explained that it's a bot... and they *still* argued with the auto-reply!

    💬 0🔄 1⭐ 0
  • 💬 0🔄 0⭐ 0
  • James Sewardjamesoff@mastodon.jamesoff.net
    Oct 26, 2026, 12:21 PM

    @darius I wrote an “AI” script for IRC bots (eggdrop) and it was just mountains of regexps and canned template sentences with a bunch of logic to let it learn words it saw and counters for “mood” and “friendship” etc and people really wanted it to be actual AI.

    I had someone argue with me when I told them they were just talking to a bot that it couldn’t be because it was making typos, but that was just an output processor which flipped chars…

    💬 1🔄 1⭐ 0
  • 💬 0🔄 0⭐ 0
  • suzannecshlan@dawdling.net
    Oct 26, 2026, 2:59 PM

    @darius
    I think the leaders of those companies are also some of the desperately hungry.

    💬 0🔄 0⭐ 0