• brucethemoose@lemmy.world
    link
    fedilink
    arrow-up
    2
    ·
    edit-2
    8 hours ago

    Agenic access is pretty routine now. LLMs use tools, and many use more dedicated UIs than a browser.

    Personally I don’t like them doing anything unless it’s in a hard sandbox.


    Also, many LLM interfaces pull more into the context than you think, secretly. ChatGPT is particularly notorious for this. For instance, when helping a family member, it referenced private medical info from another chat it absolutely should not have, unprompted. Hence for some things, I like to use “raw” completion interfaces where I know precisely what’s going into the context.