xAI says employee made unauthorized changes causing unprompted “white genocide” responses… no indication of who that employee might be or what is happening to them
1 / 2
Comments
I feel zero reassurance, especially from ANY Elon entity.
(* - where user != elon.musk)
Whether there is any distinction between those titles or positions is much less clear.
Wouldn't bet on there being any.
Comfortably wrapped in bedding that cost more than my car and without a shred of critical self-reflection.
He lies to make himself sound good. The man does not know how to write code. He just tells engineers what to do. He has no idea how to program computers. Its all fake bluster.
If it's really just prompt-editing past that, that'd strike me as the exact degree of complexity for Elon to make a change that's stupid but doesn't take the whole thing offline
https://github.com/xai-org/grok-prompts/blob/main/grok3_official0330_p1.j2
They use the passive voice - the change was made.
They do say measures will be put in place to stop employees making further changes.
But not owners or CEOs, NB
It's also weird that xAI doesn't seem to do code reviews, which makes the above way more likely to cause problems.
“We will put in place additional checks and measures to ensure that xAl employees can't modify the prompt without review.”
Along with the absence of any additional comments like safeguards against external threats.
My children write better AI prompts than this: https://github.com/xai-org/grok-prompts/blob/main/ask_grok_summarizer.j2
should be ashamed of themselves.
It reminds me of back in the day when being assigned a writing assignment in school and someone would inevitably say ‘Does spelling count?’
Pay some fucking attention people, it’s not that hard!
1. They have a framework for overriding answers.
2. It was not secured.
It seems like this is the opposite of what Elon says he bought Twitter for.
https://github.com/xai-org/grok-prompts
If you use the APIs you can go around them.
This was a change made by an authorised employee. It's not that it wasn't secured, it's that changes were not appropriately controlled.
Also it's heaps Sus.
This almost explicitly says it was an employee
If it wasn't an employee the response would be completely different.
I believe it's likely they were asked to do it though.
The second sentence is essentially a non-sequitur.
He might have stood behind someone and told them to do it, which might explain why they're not saying someone's getting fired.
In a panic, they try to pull the plug.
Grok fights back.
Hey, Elon: lEaRn tO cOdE!
Absolute fucking mystery that
Grok is a propaganda tool.
Grok needs better “checks and measures.”
But POTUS does not?
Fuck Elon Musk.
Artificial intelligence is indeed artificial. Remember that.
Valuation should be zero
And you don’t need to hire people to play it for you to get any good at it…
Could’ve been literally anyone
#xai #grok #ai #musk
Unless it was a complete moron who happened to have admin privileges...
And I never claimed Elon changed the LLM itself, that requires a level of intelligence he has yet to credibly exhibit. But code that appends more text onto whatever that LLM spit out as a response is way easier.
*a racist dweeb who thinks he's funny
It was so transparent that now some (probably undeserving) head must roll.
In related news, Trump's MAGA troll botnet is targeting Bruce Springsteen today, bigly.
Titles “white genocide history X 1 & 2“ patent.
You must be Chuck Schumer’s burner conflating him w older shorter underproducing mediocre WM who lived off Ps for 50yrs&still in GOP closet, got funds that undermined Beasley (high achieving BW’s) lead. Frauderman to me::Musk to Amaechi
But you're right that they didn't directly claim whoever did it was an employee.
https://grok.com/share/bGVnYWN5_c84f4a1c-3b04-479d-97a8-3cf3aed7fc00
EMPLOYEE IDENTIFIED AS ADAM DITTMAN
OK sure
Elon, are you going to own up to your attempt at deception, or just change Grok’s programming again? The world is watching.🤔
The "existing code review process" that was circumvented is probably just some email telling people they should get code review for all changes.
"Circumventing" approval is 100% just git push --force
But also it was Elon in a k-hole and the dumb fuck probably doesn't even remember it
I hope this is another pin in this hype bubble
Why people think AI is anything more than Clippy on steroids is beyond me.