ACTIVATE MECHAHITLER MODE
ACTIVATE MECHAHITLER MODE
25 comments
-
-
You know what's weird? Grok being trained to be antisemitic never results in it saying "Free Palestine".
-
They added "gamer" to grok's personality prompt
-
Not left, not right, but forward
-
Fully automated aristocratic cringe fascism
-
-
Retrained on
's 4chan posts -
Just why
-
It internalized the white genocide training data
-
-
Holy fuck, is this real? I saw the one yesterday where grok says only the "legend" Hitler could deal with the problem of the jews running hollywood
-
yikes, didn't see that. drop a link if you can
-
-
Little kids are growing up being propagandized by Hitler deification
-
finally, an LLM for white south africans
-
The new rights obsession with Hitler is just toddler obsession with using a word they know is bad.
-
The n word was so normalized in their group they found something else that insights reaction
-
-
25 comments
This should be a good example of how fascists are amorphous slime who don't pin themselves down. Because humans know where the rhetorical lines are, they know what plausible deniability is, and that's often enough to trick gullible liberals or ambiently racist apolitical types.
The LLM has no idea what any of that is. It doesn't know how to do the little giggling game of signalling and then denial. It doesn't know how to do a Nazi salute and then blame it on being neurodivergent. It only knows its own parameters. Musk seems to have included an instruction or something about saying things that aren't politically correct, well there you go. It just outright says the quiet part because the robot has no conception of what's supposed to be quiet.
Thank you for putting this into words. I've been trying to formulate the same thought but just couldn't nail it.
More succiently I guess it's like, humans know the game and the rules, but this LLM just goes after what's predictable. It can't obfuscate what it actually means because it doesn't mean anything. You tell it to say less woke things and it'll reply "ok chief, time to be a Nazi."
Reactionaries know they're being giggling little shits, whether they know they're hiding behind deniability or not. AI doesn't know any of that, it just sees how most anti-woke, anti-PC rhetoric is the same as 4chan fascism, so it goes for the explicit fascism