what can language model embeddings tell us about whales speech, and decoding ancient texts?      (on The Platonic Representation Hypothesis and the idea of *universality* in AI models)

The title is slightly misleading, it’s less that all AI models are the same now, and more that they’re converging. The more we improve models, the less variation there is.

  • supersquirrel@sopuli.xyzEnglish
    7·
    12 days ago

    When you keep mixing pretty colors together it always inveitably ends in a boring brown muddy color.

    The hilarious thing is big tech is unable to grapple with this, they existentially need mixing more and more colors to create more and more vibrant colors that only special huge corporations that criminally stole each of our individual hues can forge… and yet it is clear for the world to see that Big Tech keeps showing us a pallette of muddy browns and exclaiming “Look at the rainbow of colors!”.