• Prox@lemmy.world
      link
      fedilink
      English
      arrow-up
      2
      ·
      3 days ago

      I guess, though I’m pretty ignorant as to how RLVR would fix the issue that arises from new coding languages or even new major versions. I’m not sure how LLMs would ever get to a correct answer if they don’t have good reference material to start from or reference.

      • General_Effort@lemmy.world
        link
        fedilink
        English
        arrow-up
        1
        ·
        3 days ago

        The assumption seems to be that an LLM can’t figure out a manual or source code. If it can’t, then you have to pay people. But that’s not a universally valid assumption.