• JakenVeina@midwest.social
    link
    fedilink
    English
    arrow-up
    23
    arrow-down
    1
    ·
    4 days ago

    This paper from 2025 has its fun bits

    Psh, last year? You can’t judge anything based on what it was doing last year. Obviously, this entire article is worthless.

    /s

    • Hirom@beehaw.orgBanned from community
      link
      fedilink
      English
      arrow-up
      3
      arrow-down
      3
      ·
      3 days ago

      In earnest, citing 1 year old researsh seems reasonable.

      It’s more important to focus on more fundamental problems with LLMs, ie environnement cost, loss of skill, plagarizing of online content and agressive scrapin, … Less so on LLM and agents unreliability as technical fixes might improve reliability to a point where it’s no longer a problem. Frontier LLMs models are reportedly reliable enough for some coding tasks given supervision.