A two-hour interview is enough to accurately capture your values and preferences, according to new research from Stanford and Google DeepMind.
    • BertramDitore@lemm.eeEnglish
      8·
      1 month ago

      Ugh someone recently sent me LLM-generated meeting notes for a meeting that only a couple colleagues were able to attend. They sucked, a lot. Got a number of things completely wrong, duplicated the same random note a bunch of times in bullet lists, and just didn’t seem to reflect what was actually talked about. Luckily a coworker took their own real notes, and comparing them made it clear that LLMs are definitely causing more harm than good. It’s not exactly the same thing, but no, we’re not there yet.

            • Imgonnatrythis@sh.itjust.worksEnglish
              3·
              1 month ago

              Have you seen current doctor visit note summaries? The bar is pretty low. A lot of these are made with conventional dictation software that has no sense of context when it misunderstands. Agree the consequences can be worse if the context is wrong, but I would guess a well programmed AI could summarize better on average than most visit summaries are currently. With this sort of thing there will be errors, but let’s not forget that there ARE errors.

      • Patch@feddit.ukEnglish
        3·
        1 month ago

        I hosted a meeting with about a dozen attendees recently, and one attendee silently joined with an AI note taking bot and immediately went AFK.

        It was in about 5 minutes before we clocked it and then kicked it out. It automatically circulated its notes. Amusingly, 95% of them were “is that a chat bot?” “Steve, are you actually on this meeting?” “I’m going to kick Steve out in a minute if nobody can get him to answer”, etc. But even with that level of asinine, low impact chat, it still managed to garble them to the point of barely legible.

        Also: what a dick move.

      • Phoenixz@lemmy.caEnglish
        3·
        1 month ago

        You just have to love that these assholes are so lazy that they first use an LLM to write their work, but then are also too lazy to quickly proof read what the LLM spat out.

        People caught doing this should be fired on the spot, you’re not doing your job.

  • Telorand@reddthat.comEnglish
    17·
    1 month ago

    Imagine sitting down with an AI model for a spoken two-hour interview. A friendly voice guides you through a conversation that ranges from your childhood, your formative memories, and your career to your thoughts on immigration policy. Not long after, a virtual replica of you is able to embody your values and preferences with stunning accuracy.

    Okay, but can it embody my traumas?

      • Echo Dot@feddit.ukEnglish
        1·
        30 days ago

        If I can make a version of me that likes its job then that will be a deviation from the template that’s worth having. Assuming this technology actually worked, an exact digital replica of me isn’t particularly useful, It’s just going to automate the things I was going to do anyway but if I was going to do them anyway they aren’t really a hassle worth automating.

        What I want is a version that has all of my knowledge, but infinitely more patience and for preference one that actually understands tax law. I need an AI to do the things I hate doing, but I can see the advantage of customizing it with my values to a certain extent.

  • kat_angstrom@lemmy.worldEnglish
    16·
    1 month ago

    a virtual replica of you is able to embody your values and preferences with stunning accuracy.

    I’m calling BS on this one. “Values and preferences” are such a far cry from Actual Personality that it’s meaningless. Just more LLM hype