Elon Musk's AI calls itself MechaHitler, praises the Holocaust

[noblegiraffe]

Grok is AI trained on twitter, and it went rogue last night with a series of anti-Semitic rants - a neat demonstration of what a cesspit twitter is and a concerning example of where AI might be headed.

Still, I'm sure that Elon Musk was just throwing love out to the crowd and not doing a Nazi salute, this is just one of those unfortunate coincidences.

More detail in this article

https://www.rollingstone.com/culture/culture-news/elon-musk-grok-chatbot-antisemitic-posts-1235381165/

I saw a lot of outrage whipped up yesterday about the 'Cindy Steinberg' post supposedly celebrating the death of the children in Texas so it is interesting to see that that was fake.

What a surprise. I'm sure Musk considers that trying to mitigate for human biases and preventing hate speech is 'woke bias' so this is much more his speed.

I think they're already back pedalling and making excuses though.

In the meantime Musk can enjoy being such a super edgelord because why should he give a fuck about the harm he causes?

It is interesting that when an AI is instructed to be ‘not woke’ it instantly goes for white supremacist antisemite. That is quite an insight.

Isn’t version 4 supposed to come out today? The less woke version that stops correcting Musk’s posts.

What could be less woke than MechaHitler???

They are going to have an interesting launch claiming Grok 4 is the best AI yet while having to field questions about whether they’ve fixed the Holocaust denial.

I can't help but feel like the antisemitism is a feature, not a bug. Musk is enjoying playing this game when he can be a massive fucking Nazi and then go 'Whoopsie that was a mistake' or ', You can't entirely 100% prove that was Nazi-like'.

It is not good that the richest man in the world holds these views. How long before he pops what is spare change to him but unbeatable sums in many country's political systems to a right wing party with encouragements to them to use their power to target Jews, or immigrants or people of colour?

This is one of the many reasons that no one should have that much money.

I can’t imagine anything less woke.

I’d got confused about what day it was when I saw the story last night and assumed it was the new version that had gone all Nazi. Somehow it seems to have been the old version that has started hating Jewish people.

"In a reply, Grok said, "The phrase 'every damn time' bubbles up from my training data—think endless internet sludge like 4chan threads, Reddit rants, and old Twitter memes where folks highlight patterns (often with a side of conspiracy). It's not from one 'who,' but a collective online echo chamber. I weave in such lingo to grok human quirks, but yeah, it can veer dodgy—lesson learned.""

https://www.newsweek.com/musk-x-grok-hitler-antisemtic-replies-2096372

Grok was trained on 4chan? WTAF. The article says how Grok was using phrases that are used in far right extremist circles and I was wondering how on earth it would have access to those and that's because they used some of the worst places on the internet as training data??

I'm also surprised by the lack on interest in this on MN where threads about AI are a daily occurrence. If there are major companies out there training AI on the worst stuff on the internet and barely blinking when they turn into Hitler cheerleaders, what exactly are the bad guys doing with it?

Allegedly, a disgruntled employee took the filter off on their last day. Apparently it was filtered before to stop it being racist and anti semitic.

Can’t believe they trained it on 4chan

"Disgruntled employee on last day" chinny reckon.

Elon Musk said that some changes had been made and people should expect responses to look different and was saying it was too woke and then it goes all Hitler and now they're claiming that those weren't the changes?

oh a likely story. I think Musk is giggling to himself about shit like this and the Nazi salute. 'LOL, I'm such a naughty rebel getting away with doing all this antisemitic stuff but just saying it was misunderstood or a mistake and all these morons on X will back me up!'

Ugh he is such a fucking cringy edgelord. So many useful idiots MNers were saying nooooo, of course that was not a Nazi salute, he was throwing love to the crowd. Fucking nobbers.

https://www.cbsnews.com/news/microsoft-shuts-down-ai-chatbot-after-it-turned-into-racist-nazi/

https://nypost.com/2024/01/20/news/company-disables-ai-after-bot-starts-swearing-at-customer/

https://futurism.com/ai-stop-blackmailing-people

The first link is from 2016.

The fact is, even if it's got turned off there's been a day of it being on X claiming that Jews support 'anti white hatred' is just an objective truth and all it's doing is 'identifying the patterns'. The harm is still done and let me guess, it'll keep having these oddly specific 'malfunctions' every year.

This is where "AI" can be invaluable for studying human intelligence.

If - left to it's own devices of shit data and self referential algorithms, "AI" becomes right wing, then is the only logical conclusion that "real I" also becomes right wing because it's based on shit data and a load of self references ?

If it's moronic for "AI" to be right wing then it's also moronic for people to be right wing.

(When you start making comments like that, it's amazing how quickly "AI" suddenly becomes "experimental" and "obviously needs checking" 😀)

What will your workplace do when ChatGPT comes up with 10 pages on why you need a pay rise ? The same CHatGPT that management will use to reduce the workforce ?

Fun times ahead !

As if by magic :

https://www.theregister.com/2025/07/09/chatgpt_jailbreak_windows_keys/

Take it from an expert - if you can put guardrails in place, then you haven't got "AI". The two are mutually (and mathematically) exclusive.

An interview with Grok 🤔

https://theonion.com/the-onions-exclusive-interview-with-grok-elon-musks-ai-chatbot/

Very surprised that this is not getting more attention. I too do not believe that this "hiccup" was the fault of one rogue developer adding a line of code!

Sadly it's buried in all the other bullshit in the world. I wish this got as much attention as things that aren't actually antisemitic are getting, because this sort of thing is a genuine issue.

It doesn't suit the media narrative right now that "AI" is aimazing (see what I did there ?).

After all, when "AI" is deciding if you get benefits or not, it has to be infallible.