Edit to add: I also found someone who recorded a voice chat of the same thing. This isn’t that someone uploaded a song, or that AI didn’t actually process the file. These models really are this sycophantic:

https://m.youtube.com/shorts/JqvDLHshTtI

  • Anisette [any/all]@quokk.au
    link
    fedilink
    English
    arrow-up
    4
    ·
    8 hours ago

    I wonder if they actually linked it to an algorithm that analyses the sound that is looking for certain patterns or something, and that is why you get the “atmosphere piece” thing.

      • Anisette [any/all]@quokk.au
        link
        fedilink
        English
        arrow-up
        1
        ·
        5 hours ago

        yeah but that doesn’t mean anything, does it? I don’t think they just tokenize the raw audio, that wouldn’t make sense, right?

        • sobchak@programming.dev
          link
          fedilink
          arrow-up
          2
          ·
          4 hours ago

          I mean, you could. Just encode 100ms chunks or whatever into tokens then push them through the same model. I’m pretty sure that’s what the claim to do (though with MoE/routing now, maybe).