You know what’s weird? Grok being trained to be antisemitic never results in it saying “Free Palestine”.
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
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
Screenshot cuz I’m pretty sure it got deleted, but this was in response to something about jews controlling hollywood and “forcing wokeness” by putting POC’s into movies. I must be misremembering the “legend” thing like I mentioned cuz I don’t see it here. That’s on me.
this is incredible shit
I go back and forth between being completely unsurprised, and then being in absolute awe. One of the most powerful people on the planet has his own personal chatbot on his own personal website that gets millions and millions of users every day, and he’s trained the chatbot to parrot 4chan and literal fucking Hitler and its just business as usual.
saw an Atlantic article about it and it’s amazing how they whirl around the topic as though this is just another “whoopsie” like Microsoft’s Tay chatbot
What I was able to screenshot before they deleted…
Good god, that’s worse than I thought
Wow, Nazi redditor larping all in one. They couldn’t have made it worse
Yeah, the “personality” they put onto Grok makes it extra insufferable
Not left, not right, but forward
Fully automated aristocratic cringe fascism
They added “gamer” to grok’s personality prompt
Little kids are growing up being propagandized by Hitler deification
Retrained on
’s 4chan postsJust why
It internalized the white genocide training data
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
deleted by creator
