Cassie Heart
@cassie
all this "groundbreaking, gamechanging" research here to spend 34 seconds "thinking" about a riddle that isn't a riddle, and still fucking it up
19 replies
2 recasts
40 reactions
agrimony↑🎩
@agrimony.eth
most LLMs seem to overleverage on style rather than substance. guessing that for the most part is model bias during its training (eg most riddles in its training set would be styled as such)
1 reply
0 recast
2 reactions
Ryan J. Shaw
@rjs
yup. https://warpcast.com/rjs/0xbb68faf4
1 reply
0 recast
1 reaction