Ancestors

Toot

Written by Joshua Thayer on 2024-10-27 at 01:30

Can we just replace the term “AI” with “applied statistics” in … all cases?

=> More informations about this toot | More toots from joshua@pigeons.club

Descendants

Written by d@nny mc² on 2024-10-27 at 02:03

@joshua that is definitely what i do

=> More informations about this toot | More toots from hipsterelectron@circumstances.run

Written by Peter Ralph on 2024-10-27 at 02:43

@joshua I wouldn't call LLMs "applied stats" though - I think the intellectual development was pretty parallel. But maybe many of their uses become applied stats?

=> More informations about this toot | More toots from petrelharp@ecoevo.social

Written by Joshua Thayer on 2024-10-27 at 03:43

@petrelharp Oh that’s interesting! You’d know better than me, but my understanding is that LLMs are essentially applications of a novel Neural Network (well, novel as of Google’s Transformer paper), and I’d call an ANN a statistical data structure.

I think “applied” is doing a lot of work in my assertion (standing for lots of engineering and research), but at its core AI including LLMs is glued together statistical models, from which some maybe-useful behavior emerges.

=> More informations about this toot | More toots from joshua@pigeons.club

Written by Peter Ralph on 2024-10-28 at 03:54

@joshua Ah I see what you mean. I was thinking about NNs coming from Turing, neuroscience, computation, which is historically pretty distinct from the discipline of statistics.

=> More informations about this toot | More toots from petrelharp@ecoevo.social