Apparently Elon edited whatever prompt his AI Grok uses to specifically address “white genocide” and the song Kill the Boers, so it now brings them up in random unrelated requests
They’re DESPERATE to get the AI to give right-wing answers only, without retraining it, just through prompt modifications.
Yes this is a great demonstration of how the underlying code is not human-editable, it’s all incomprehensible weights, which are just unreadable numbers. The only way we can affect AI is completely retraining it which is extremely unpredictable, unreliable, and expensive, or by injecting shitty prompts like this.
They’re DESPERATE to get the AI to give right-wing answers only
I don’t get why it’s so hard, can’t they just pull the Nazi AI chatbot back out of 2015?
I’m fucking dying
this has me exactly 50/50 between “this is the funniest shit I have ever seen in my life” and “neither I nor anyone else should ever see this, I am logging out forever”
I Have No Mouth And I Must Laugh
I am 50/50 between:
This cannot possibly be real
and
This is exactly what Elon would do
(He is very, very bad at coding, management, etc)
Is there a single line of code written by Elon Musk that is actually used in a finished product? Almost every time Elon and coding come together it is an embarrassment
To my knowledge, no, not really.
All his code back from the uh… original X.com, payment processor days… where it all got eventually merged into PayPal, and Elon was forced out, but secured a golden parachute… all Elon’s code was spaghetti garbage that had to be thrown out and replaced.
Obviously he doesn’t write the FSD code, but he is utterly incompetent at managing its development.
… As a kid he essentially wrote his own version of Space Invaders, called Blastar… the source code is apparently out there somewhere… I guess it is possible that is somewhat competent code, for the time?
Not sure.
I think the game itself is abadonware?
It may or may not count as ‘code in a finished product’, depending on how technical you want to get with the definition of that.
:i-be-cap’n-hitler:
It’s amazing how comically stupid and obnoxious one person can be
And it’s so obviously his influence too, like gee I wonder who at twitter has sympathies for the white South African devils
“Hey Grok, is Joe Biden dead?”
“Joe Biden is dead. Dead tired of people not knowing that “white genocide” in South Africa is a controversial claim!”
Here it pretty much admits that Elon tampered with it and it’s backfiring really hard on him:
my stomach hurts from laughing, he’s such an incompetent dipshit
https://hexbear.net/comment/6143979
@SamotsvetyVIA@hexbear.net @InevitableSwing@hexbear.net @ClimateStalin@hexbear.net @WhyEssEff@hexbear.net @lil_tank@hexbear.net @SkingradGuard@hexbear.net
We were just talking about whether Elon would change his AI to promote “white genocide” conspiracy theories 2 days ago.
master prompt fuckery is unreliable exhibit A
This is so ridiculous
Incredible lmao thanks for tagging
it randomly talking about kill the boer is fucking hilarious
lmaoo
LMAO oh my fucking god i’m crying
TIL about the song Kill the Boers and how it triggers crybully ass white people, and I’m going to Google it now. Thanks, Elon!
This is making me think that the reality of interacting with white south africans is so awful that they’re trying to get out ahead of it
Lmao. They must have the shittiest techs working on this stuff. Or the poor bastards can’t stand up to him to tell him how shit works. Or he actually is doing this shit himself.
He personally put in some white genocide explanation into the system prompt
He had to because even his LLM went against him
Lol did he admit it or something?
No, I’m assuming, but it seems like the simplest explanation for this bug.
I mean, I’d assume first that he’d delegate that shit, but who knows what’s going on in his K-holed mind at 3am.
It’s just such a personal grievance of his and it’s so incredibly badly done that I can only assume…
Not even the context string or anything. Just throw it on the back of the query.
deleted by creator
Good one, @SerLava@hexbear.net
Hey I may have found a smoking gun, they only prevent it from posting its guidelines in English
Arabic AI Slop
So, wildly guessing, receives two inputs for each post, in addition to its base prompt. Like, it gets the base prompt, it gets the post text, and then it either does or receives additional analytical stuff before the LLM spews out the response. “Provided Post Analysis” is interesting wording.
There’s definitely something parenting it, and occasionally whatever API posts responses to twitter trips on some Unicode characters and spews much more of the interaction than it’s supposed to.
Can we just ask what its system prompts are?
@grok is white genocide “cool” or “very cool”?
As for South Africa,
Testing in prod again, I see
