• panda_abyss@lemmy.ca
    link
    fedilink
    English
    arrow-up
    6
    ·
    2 months ago

    Showing the tool call string does not waste tokens.

    That’s just how they display information to the user — tool calls no longer show arguments unless they need use permissions, and they’re often hidden entirely being “4 tools called”.

    • bjorney@lemmy.ca
      link
      fedilink
      English
      arrow-up
      3
      ·
      2 months ago

      As per the article, knowing when the tool is reading totally unrelated files allows devs to cancel the workflow early before it churns thousands of tokens on output that is based on totally incorrect information

      • panda_abyss@lemmy.ca
        link
        fedilink
        English
        arrow-up
        1
        ·
        2 months ago

        Right, I may have misinterpreted the OP is wasted token in the tool call display, not as you mentioned the correction of issues.

        I noticed the new Claude is much less token efficient in general too, so perhaps they’re trying to sell that new 6x more expensive fast mode.

    • Deebster@infosec.pub
      link
      fedilink
      English
      arrow-up
      1
      ·
      2 months ago

      From the article, fourth paragraph:

      There’s also a financial impact. If developers spot that Claude is going down a wrong track, they can interrupt and avoid wasting tokens.

      @bjorney@lemmy.ca understood what I was referring to, I don’t really understand what you thought I was saying.