Currently reading the MS GPT-4 paper. At the moment I remain… unimpressed… by the model’s “human like” behaviors.

That said, I’m starting to think that there may be something profound here. Just not what the authors think it is.

Reluctant anti-utopian | Breaker of things | https://necopinus.xyz

IMHO, the versatility of GPT-4 strongly suggests to me that ALL human knowledge domains may use the SAME underlying symbolic system and rules. If true, this means that all human domains involving symbolic manipulation may, in fact, be isomorphic to each other.

If true, this would be a profound epistemological leap forward: Humans have a single “map” that we have (more-or-less) successfully mapped to EVERY territory that we’ve encountered.

Every model works if you use enough epicycles 🫣

host of /degen and /houseofdegen | curator for /wakegallery | husband, father, degentleman, haberdasher 🎩🥂

To an extent. But I think the implication I’m groping for here is “why should a single model work this well at all?”

TBF, this is kind of the same observation as “The Unreasonable Effectiveness of Mathematics in the Physical Sciences”, but with its generalization kicked up a couple of (significant) notches.