Opinion: Fanboys think it's the greatest thing since sliced bread. Devs aren't nearly as won over

Some people seem to think writing a C compiler is hard work. Well, yes, if you’re writing it in assembly language, it is. But I hate to break it to you: Computer science undergraduates write C compilers every semester. Heck, you can learn to do it yourself thanks to Daniel McCarthy’s Developing a C Compiler From Scratch online class. And the class will cost you a lot less than the $20,000 it cost Anthropic.

  • arcine@jlai.lu
    3·
    1 day ago

    As the article says : a C compiler without assembler nor linker is a very standard undergraduate project.

    Good students are supposed to be able to do it alone, average students in groups of around 4.

    So for AI that’s supposedly “PhD level in everything” ; this shouldn’t be considered that impressive, even less so with 16 of them on the case…

    Seems like even they know their claims are super overstated.

    • cecilkorik@lemmy.caEnglish
      7·
      4 days ago

      Plagiarism machine turns out to be tolerably competent at plagiarism. Whenever you need some quick plagiarism, go ahead and grab the plagiarism machine, it’ll do an “OK” job!

    • frunch@lemmy.world
      3·
      4 days ago

      I don’t know who that guy in the gif is, but he looks like he’s from the same exclusive gene pool as Jake and Logan Paul

  • hendrik@palaver.p3x.deEnglish
    151·
    4 days ago

    it was given "highly intricate and thorough test suites along with execution harnesses, all crafted by humans, with the harness designed specifically for the AI’s use.

    the system was developed using the very code base it is supposed to replicate.

    I feel it is a bit impressive, how it’s able to write coherent code. And larger code bases and make the puzzle pieces align. That’s some achievement, I guess? But it’s the same issue as always. We keep proving AI can reproduce existing things. Do mashups and recombine. Within tight constraints.

    We do not, however, see it create or invent new things. I mean why not use $20k to develop something useful, which isn’t already there? Like a modern successor to E-Mail? That’d be badly needed. A great showcase of reasoning abilities… Useful…

    I think because it can’t do it. And I’m not sure if that’s just a performance issue. It’s like evolving from a Ctrl+C, Ctrl+V Stackoverflow programmer to a proper one. But that’s not easy. It takes years of study. And takes developing a solid understanding of what you’re doing, plus intelligence. And maybe coding agents won’t just automatically transform into that. Despite the claims.

    (Edit: Strg -> Ctrl)