I’ll return to my “ergodic hypothesis” and assume, contra “grammaticalist” assumptions regarding the endless possibilities of uttering new sentences, that language, in fact, encloses within a limited system of idioms—of course, on one level, we are always saying things that no one has ever said before but it is always by recombining elements of the system. Just like we have never left the originary scene or moved beyond ritual (merely replacing desecrating rituals for consecrating ones) we have never moved outside of the oral. “Ergodism” counters one of the main attacks on Large Language Models, i.e., that they just recombine existing language—yes, of course, but if you want to differentiate them meaningfully from the human on those grounds you’d have to distinguish between different modes of recombination, not simply contrast recombination with “originality.” Most important for me here is that the ergodic hypothesis, as opposed to the grammaticist, allows one to think in terms of working on language rather than “clothing” thoughts in language.
Designing Idioms for Data Treatment
Designing Idioms for Data Treatment
Designing Idioms for Data Treatment
I’ll return to my “ergodic hypothesis” and assume, contra “grammaticalist” assumptions regarding the endless possibilities of uttering new sentences, that language, in fact, encloses within a limited system of idioms—of course, on one level, we are always saying things that no one has ever said before but it is always by recombining elements of the system. Just like we have never left the originary scene or moved beyond ritual (merely replacing desecrating rituals for consecrating ones) we have never moved outside of the oral. “Ergodism” counters one of the main attacks on Large Language Models, i.e., that they just recombine existing language—yes, of course, but if you want to differentiate them meaningfully from the human on those grounds you’d have to distinguish between different modes of recombination, not simply contrast recombination with “originality.” Most important for me here is that the ergodic hypothesis, as opposed to the grammaticist, allows one to think in terms of working on language rather than “clothing” thoughts in language.