Need to let loose a primal scream without collecting footnotes first? Have a sneer percolating in your system but not enough time/energy to make a whole post about it? Go forth and be mid: Welcome to the Stubsack, your first port of call for learning fresh Awful youāll near-instantly regret.
Any awful.systems sub may be subsneered in this subthread, techtakes or no.
If your sneer seems higher quality than you thought, feel free to cutānāpaste it into its own post ā thereās no quota for posting and the bar really isnāt that high.
The post Xitter web has spawned soo many āesotericā right wing freaks, but thereās no appropriate sneer-space for them. Iām talking redscare-ish, reality challenged āculture criticsā who write about everything but understand nothing. Iām talking about reply-guys who make the same 6 tweets about the same 3 subjects. Theyāre inescapable at this point, yet I donāt see them mocked (as much as they should be)
Like, there was one dude a while back who insisted that women couldnāt be surgeons because they didnāt believe in the moon or in stars? I think each and every one of these guys is uniquely fucked up and if I canāt escape them, I would love to sneer at them.
(Semi-obligatory thanks to @dgerard for starting this - gonna try posting last weekās thread a different way this time)
The first half was OK, but then they cited this paper.
I havenāt read the paper, and probably wonāt, but what the shit is this?
gave it a quick skim. I lack any relevant background. the bit they push most seems to be that you can improve the performance of error detection tools by determining the most important tokens in an answer and running your tools on the tokens near those. this seems to be in contrast to absurdly naive approaches like averaging the tokens (???) or just looking at the last token of the response (???).
what are the most important tokens? theyāre the ones that change the factuality of the answer if you change them. how do you determine that? you donāt, lmao. you just ask an LLM what the most important words are
what are the error detection tools? you will never guess
It turns out if you can just make the machine know what the truth is and say that you donāt get hallucinations. Unfortunately the truth isnāt emergent from pure language models and expressing Truth through language alone has been something challenging the human race since Krog try to teach Torg how make stick but pointy.
Derrida look intensifies
tired: checking incoming packets for the evil bit
wired: checking LLM outputs for the truthfulness bit