myrmepropagandist on Nostr: npub1l0xn8…9ckv7 npub184hne…k4shr npub15dwvs…jkg2d That's part of it, but there ...
npub1l0xn8scmlds3xya4wk3vw20nfhz2srcxf7e8phtharmk8rj98ucq69ckv7 (npub1l0x…ckv7) npub184hne09jyjfmh252gcqlhqughgdpdmxkrxs9annt5vp54ah4t5aqnk4shr (npub184h…4shr) npub15dwvsprvd5wzsx3se0plmuyuw7uc386m496pgntp8v44wpl377hqpjkg2d (npub15dw…kg2d)
That's part of it, but there are things about the ways that humans (and corvids!) process sound and words that I think are more basal and we could learn from that and make better models?
LLM don't really have anything but statistics. It's just raw probability and reliance on the repetitiveness of language. If words cluster in one way most of the time, doing so again will seem like it's meaningful communication.
What if this system was less lazy?
That's part of it, but there are things about the ways that humans (and corvids!) process sound and words that I think are more basal and we could learn from that and make better models?
LLM don't really have anything but statistics. It's just raw probability and reliance on the repetitiveness of language. If words cluster in one way most of the time, doing so again will seem like it's meaningful communication.
What if this system was less lazy?