• maria ~@lemmy.blahaj.zone
    link
    fedilink
    English
    arrow-up
    0
    ·
    edit-2
    3 months ago

    yesss its a weird usecase. its easy to think ur getting so much out of it while theres actually rather little to gain.

    (okay fine yes im just commenting on this post cuz its got more likes than mine… and its a screenshot! >o< grrrrr- )

    if u ask mister gpt or daddy claudius for a web search thing, you lose out on the actual sources. Which sucks cuz they are right there in the LMs context, but u dont get to see it. u dont get to see the sources.

    this requires a mockup. imma be back in a few hours or so and post a mockup on Qwen community and then link it here. So here u go: have this pointless indicator showing that im working on something, maybe:

    👩‍🍳 Cooked for 1 hour and 20 minutes

    okay i cooked. 🍳

    the post is here and if u dont care to open a link but do care to click a dropdown, here is the mockup:

    mockup i made

    mockup showing conversation where assistant highlights the source instead of unverifiably reguaritating the content

    the idea is that the LM highlights the actual sources by writing special syntax which shows up in the UI as the sources themselves. Here an example.
    The LM writes:

    Here is that part of the script:
    <show>
    <path=“res://script.gd”>
    <from_line=7>
    <to_line=7>
    </show>

    This gets parsed and shows up as this:


    Here is that part of the script:

    var whatever: String = "the content of this file is"
    

    soooo… it doesn’t regurgitate, but highlights the actual sources. The example is about code but this is easily extended to regular web search. its just text so… its the same thing.

    sigh i spend too much time on lemmy.

    EDIT: i moved the post from qwen community (which is on blahaj zone) to FOSAI which is not on blahaj zone cuz i feel this stuf doesnt belong here.