• PhilipTheBucket@ponder.cat
    link
    fedilink
    English
    arrow-up
    2
    ·
    8 days ago

    Converting from one format to another, it can do like gangbusters. I wouldn’t trust it to summarize stuff from its training data, it can do a little bit better with summarizing stuff you give it, but just mechanically finding the text and putting it verbatim into a different markup it’s pretty capable with.

    • ComradeMiao@lemmy.dbzer0.com
      link
      fedilink
      English
      arrow-up
      3
      ·
      8 days ago

      Even reformatting has caused me issues. My best example is I gave it 100 citations in a non standardized format and asked for MLA. It returned 100 in MLA but randomly 10 of the books were made up. It decided to delete ten I sent at random and make them up instead of just giving me what I sent

      • PhilipTheBucket@ponder.cat
        link
        fedilink
        English
        arrow-up
        2
        ·
        8 days ago

        Oh… yeah, you might have a point. Beyond a certain size of repeated things, it sometimes goes haywire, I’ve seen that.

        • ComradeMiao@lemmy.dbzer0.com
          link
          fedilink
          English
          arrow-up
          2
          ·
          8 days ago

          I didn’t consider length! That’s a good point too

          Another goofy example is I asked it a python question using a specific package import . I sent a big chunk of code. It answered using a package I wasn’t even importing breaking everything. It could never figure it out either lol

          • PhilipTheBucket@ponder.cat
            link
            fedilink
            English
            arrow-up
            2
            ·
            8 days ago

            Yeah, that kind of thing requires reasoning, and it goes awry almost immediately. It’s still pretty useful for generating snippets of boilerplate or finding stuff in big chunks of code, but I more or less gave up on having it actually create anything nontrivial in code.