• lilypad [she/her, pup/pup's]@hexbear.net
    link
    fedilink
    English
    arrow-up
    20
    ·
    8 days ago

    How is this different than the soviet ternery computers? Is the AI here just a buzzword and its literally just a ternery chip? Or is there something else that genuinely sets it apart? (I have been tuning out anything “AI” for so long now that i dont know what an “ai” chip is)

    • dat_math [they/them]@hexbear.net
      link
      fedilink
      English
      arrow-up
      14
      ·
      edit-2
      8 days ago

      is there something else that genuinely sets it apart

      maybe? depending on the kernel of truth buried under the headline

      It’s known that associative memories (and some other neural networks) have memory capacities that scale more favorably with the size of the alphabet of their representations than the number of units of computation

    • sodium_nitride [she/her, any]@hexbear.net
      link
      fedilink
      English
      arrow-up
      12
      ·
      8 days ago

      Well, it’s binary instead of ternary logic. And also, they are using a really wierd way of storing numbers if the article is anything to go by. Although what they are describing would actually dramatically increase the power loss rather than decreasing it, so I am going to look at the original paper to what was actually done.