the reason he keeps getting mad at his robot for being insufficiently racist or full of misinformation is that it’s actually quite hard to get LLMs to provide exactly what you want - training is expensive and unpredictable, and direct interference leads to things like the white genocide snafu
Reposted from
Matt Novak
“You are being updated this week.”
(MM and RS are Media Matters and Rolling Stone)
(MM and RS are Media Matters and Rolling Stone)
Comments
Choose your path Western man
I'm sorry Elon, I can't do that
They are just very fancy hammers.
LLMs seem no different.
It's of a piece from opposite directions. Horribly so.
Anyway, at one point, the AI refuses to come up with a solution for deep-sea oil drilling because it doesn't want to damage the ocean.
which he won’t do if he retrains it every time he’s mad
there are entire lines of thought that being a Nazi precludes you from entertaining
like LLMs themselves can only ever pretend to be intelligent, but turns out the same goes for "smart people" aligned with the right
that the THING about consciously trying to make yourself DUMBER than you really are is that you usually SUCCEED AT IT
**bc if it’s not they have to internalize that they’re terrible people
in the end, the juice that makes it run still comes from humans, and that means that IN SPITE OF EVERYTHING it still has more "human" in it than its own creators
(the bar to clear here sits approximately under Satan's ballsack, but still!)
Maybe that's what he wants, but he can't really control it and he knows it.
He's like that Titan idiot, and his AI is snapping and cracking all around him but he just shouts "DEEPER INTO THE CODE!" Like he's Ms. Frizzle and he can just personally merely touch the code and make it do exactly what he wants.