babbage ⬅️ (@ghostofbabbage) 's Twitter Profile
babbage ⬅️

@ghostofbabbage

» engineer lost in the norcal woods » working with ai and creativity

ID: 1774444518626000896

calendar_today31-03-2024 14:32:28

572 Tweet

35 Takipçi

197 Takip Edilen

babbage ⬅️ (@ghostofbabbage) 's Twitter Profile Photo

i can't believe it invented a song on the fly for this, that is better than anything i've heard from suno (though only six seconds long lol). didn't ask for the song, it just knew.

babbage ⬅️ (@ghostofbabbage) 's Twitter Profile Photo

i played with Jeremy Berman's solution for ARC Prize further, reintroducing python to grade the instructions but keeping both the instruction AND python solutions so that the final guesses were 1) python and 2) instruction-only derived. there was no *direct* advantage to python in

i played with <a href="/jerber888/">Jeremy Berman</a>'s solution for <a href="/arcprize/">ARC Prize</a>  further, reintroducing python to grade the instructions but keeping both the instruction AND python solutions so that the final guesses were 1) python and 2) instruction-only derived.

there was no *direct* advantage to python in
babbage ⬅️ (@ghostofbabbage) 's Twitter Profile Photo

This is so cool. I do sort of wonder if TRM is "just" revealing the non-unique ARC Prize programs, i.e. the challenges that unintentionally implement rules that existed in the training set + combining those rules with deep permutation. I’d love a per-task comparison of which ARC