Login
You're viewing the discuss.systems public feed.
  • Spooky Simonsonthedansimonson@lingo.lol
    Oct 6, 2026, 1:56 PM

    RE: mamot.fr/@pluralistic/11739412

    Cory Doctorow's morning thread helped make something click I'd been wrestling with for a long time but struggled to express.

    When I got into linguistics and AI about 15 years ago, I was imagining devising systems that made it easier to express expertise.

    What I found was techniques that barreled toward what we have now -- statistical inference on sets of data that produced an approximation of the phenomenon under scrutiny.

    It was a bit of a disappointment, tbh

    💬 1🔄 2⭐ 2

Replies

  • Spooky Simonsonthedansimonson@lingo.lol
    Oct 6, 2026, 2:00 PM

    That's not to say the choice toward statistical inference wasn't motivated. Our theories of the world gloss over a lot more than we consciously bring to light, and statistical methods patch over a lot of those gaps.

    But I always imagined, through understanding of language, you could better understand how to express ideas in a way that was theoretically elegant, with inference filling the gaps where theory could not, perhaps with analysis of those gaps advancing theory.

    💬 2🔄 1⭐ 2
  • Spooky Simonsonthedansimonson@lingo.lol
    Oct 6, 2026, 2:06 PM

    I still struggle to express exactly what I was going for there -- I feel like I've had moments where I was closer to it, but linguistics itself is still a relatively primitive science and so remains a relatively limited guide. A lot of work has been put into the field, but it's very easy to stumble into phenomena where no one has looked because language is so vastly complicated.

    💬 1🔄 0⭐ 1
  • Spooky Simonsonthedansimonson@lingo.lol
    Oct 6, 2026, 2:10 PM

    That said, CD's thread really clarified something there -- I was looking to enhance expertise, but the LLM companies are looking to displace it.

    And that's one of the things that's been baseline pissing me off the last 5 years.

    💬 0🔄 0⭐ 0
  • Denny Vrandečićvrandecic@mas.to
    Oct 6, 2026, 2:23 PM

    @thedansimonson do you think that formal languages would be a better fit for knowledge representation than natural languages? Do you think that we can learn something from natural languages about knowledge representation that we're still missing from formal languages? I'm struggling in that topic too.

    💬 1🔄 0⭐ 0
  • Spooky Simonsonthedansimonson@lingo.lol
    Oct 6, 2026, 2:49 PM

    @vrandecic it depends on what we're trying to do. when we need precision, formal languages are necessary, but we often lack the self-understanding of how imprecise our actual knowledge is.

    💬 0🔄 0⭐ 0