• 7K Posts
  • 1.15K Comments
Joined 11 months ago
cake
Cake day: December 11th, 2024

help-circle






















  • Not really surprising that they’re good at analyzing language, since they are Large Language Models after all. Still neat to see, though. Here’s the most interesting bit:

    In the phonology task, the group made up 30 new mini-languages, as Beguš called them, to find out whether the LLMs could correctly infer the phonological rules without any prior knowledge. Each language consisted of 40 made-up words. Here are some example words from one of the languages:

    • θalp
    • ʃebre
    • ði̤zṳ
    • ga̤rbo̤nda̤
    • ʒi̤zṳðe̤jo

    They then asked the language models to analyze the phonological processes of each language. For this language, o1 correctly wrote that “a vowel becomes a breathy vowel when it is immediately preceded by a consonant that is both voiced and an obstruent” — a sound formed by restricting airflow, like the “t” in “top.”

    The languages were newly invented, so there’s no way that o1 could have been exposed to them during its training. “I was not expecting the results to be as strong or as impressive as they were,” Mortensen said.

    I’ve also tried out various LLMs on daily puzzles that it couldn’t have been trained on, like Connections and it does a really good job. I don’t think that the end of humanity is nigh or anything dramatic like that, but IMO this invalidates people that really want to hate AI and claim has 0 intelligence.











  • The takeaway:

    Complexity is not a virtue.

    Start simple. Add complexity only when you have proof you need it.

    Out of the Tar Pit is a classic paper on this topic as well. It goes into a lot more depth, but can be summed up as:

    Simplicity is Hard

    OTOH, I feel like I should like Go because it’s “simple”, but it ends up pushing down on the waterbed of complexity only to have it pop up elsewhere. Kind of like how brainfuck is about as simple of a language as you can make, but there’s a reason it’s a toy language.



  • Yeah, bot posts are good IMO for things that are like once per day or less. More frequently and it can end up being spammy. OTOH, I’ve browsed all a few times and seen interesting posts from HN that I then went over and read, so there’s something to be said for that too. Maybe if there was a more nuanced option, like “don’t show me bot posts unless they’ve been upvoted by a non-bot” or something like that. Or maybe if bots reposting from elsewhere did that filtering beforehand, like top 10 per day or something




  • m_‮f@discuss.onlineOPtoPeanuts@midwest.social1951-11-01
    link
    fedilink
    English
    arrow-up
    2
    ·
    9 days ago

    It’s also unusual that he pluralizes the first part, looks like that was more common at first, in the early 20th century:

    https://www.sootoday.com/rooted/origin-of-phrase-trick-or-treat-in-print-traced-to-the-sault-6033664

    According to Barry Popik, an American etymologist from New York who has been studying language for decades, the Sault is the first place to reference ‘tricks or treats’ in print anywhere in the world.

    The citation comes from an edition of The Sault Daily Star on November 1, 1917, which detailed a brief story about local youngsters who were out celebrating Halloween the previous evening.

    “Tricks or treats you could hear the gangs call out,” the article read.

    Sounds like there was more regional variation early on and things were still in flux when this comic was made. The references I could find to “money or eats” were from the Midwest, so probably a variant that died out in favor of just “trick or treat”