Grok's white genocide fixation caused by 'unauthorized modification'
theverge.com
theverge.com
1 - It was some super valuable 10x guy
2 - (more likely) it was Elon Musk
Did Google mention firing the dev(s) who blocked Gemini from generating photos of white people?[2]
Most likely no people were fired in either of such cases because they were only following orders from above congruent to the company's internal political and cultural biases, or if they were acting rogue, they got hefty severance packages in exchanged for signing NDAs not to talk to the press about the toxic and possibly illegal things going on inside the company.
But either way, no company wants to publicly talk about firing rogue workers since its bad press no matter how you slice it, plus its an admission of guilt of company's culture being rotten or even illegal behind the scenes. They just deny and call it a bug then stay quiet while changing things behind the scene till people forget about it.
[1] https://www.nytimes.com/2025/02/25/technology/iphone-dictati...
[2] https://timesofindia.indiatimes.com/etimes/trending/googles-...
Musk has a history of doxxing.
Pichai/Cook does not have a history of doxxing.
Ergo, we would expect Musk to doxx somebody given the opportunity while we would not expect the same of Pichai/Cook. The standard being, somebody who likes to doxx will doxx.
2. Apple's case is more similar, but their dictation feature is not a core product in the way that xAI's chat bot is. In other words, you'd expect more checks to ensure that something like your system prompt can't just be modified by some employee (on purpose or inadvertently).
Why does that flaw only have a bias against white people? It's not like it's an accident that nobody inside Google noticed before release. That prompt bias was put there by employees at Google.
Think of it the other way, if Gemini had a bias against non-white demographics, would you still brush it off as just a flaw, or as a form of bias/discrimination of employees at Google who code, test and approve this stuff?
Remember when Google's image recognition mistakenly took a photo of two black people and labeled as monkeys? They had hell to pay for that mistake yet if it refuses to acknowledge white people then it's just a innocent whoopsie.
- AI companies not training their algorithms on sufficiently diverse data, not tuning their algorithms sufficiently to penalize bias, and not testing it sufficiently to ensure the responses are not biases.
- Someone at an AI company deliberately modifying the system prompt in order to encourage responses of a given type.
Not saying the latter is worse than the former, but they're completely different problems. The xAI problem, in the case, was the latter.
The team responsible for training and alignment did a remarkably good job at being impartial. If it wasn't for that we might have fewer incidents of "rogue employees" messing with the prompt
https://x.com/i/grok/share/Nj2tsvCpgEfU3OCHh0Ci4qHTf
Details here: https://www.euronews.com/my-europe/2025/03/03/is-ai-chatbot-...
That seems like the kind of pseudo-socialist red tape that blocks 100x engineers from getting things done.
If you were responsible for the releases of your flagship chat bot, how many layers of control do you think you would have over the system prompt, arguably its most important (and potentially damaging) component?
Either:
1. There was no rogue employee.
2. xAI doesn't know how to ship production code.
This happened with Grok saying that Elon Musk & Trump were disinformation spreaders. Here is Grok giving outs its system prompt fix for that "issue":
https://x.com/i/grok/share/Nj2tsvCpgEfU3OCHh0Ci4qHTf
https://www.euronews.com/my-europe/2025/03/03/is-ai-chatbot-...
{%- if dynamic_prompt %}
{{dynamic_prompt}}
{%- endif %}What we’re going to do next:
- Starting now, we are publishing our Grok system prompts openly on GitHub. The public will be able to review them and give feedback to every prompt change that we make to Grok. We hope this can help strengthen your trust in Grok as a truth-seeking AI.
- Our existing code review process for prompt changes was circumvented in this incident. We will put in place additional checks and measures to ensure that xAI employees can't modify the prompt without review.
- We’re putting in place a 24/7 monitoring team to respond to incidents with Grok’s answers that are not caught by automated systems, so we can respond faster if all other measures fail.
The trust is already broken. They can claim they will open source the system prompt all they want but there's no point in believing what they say. Elon clearly does what Elon wants to do.
I am having to comment and share my opinion on all of the posts separately because someone or some group of people keep flagging the posts to try and hide discussion about this very real incident.
https://news.ycombinator.com/newsguidelines.html
And to be fair, I don't think me sharing the same comment across three separate HN posts (two of which are flagged, so invisible to users unless they search for it) counts as spamming duplicate content though.
HN is a spirit-of-the-law place, not a letter-of-the-law place anyhow (https://hn.algolia.com/?dateRange=all&page=0&prefix=false&qu...).
Now we have a sizable contingent of posters who have decided that some stories are too dangerous for open discussion. This surprises me is that there is no large scale effort to fight back against these 'flagged' topics. Where have our free speech fighters gone?
Let us be honest in that they really only believed "free speech more me but not for thee".
Moderators and longtime top contributors to HN come to recognise that these regular dramas about politics and culture wars bring out the worst in people, and lead to the very worst discussion threads we see on HN, precisely because “everyone has lots of feelings” about them.
It’s not wrong for people to have strong feelings about these topics. They're important issues, we get that, and that's why we try to be at least somewhat accommodating of discussions about them.
But when they routinely turn out to be the worst discussions we ever see on HN, and start to turn HN into something very different from what it is intended to be, we think that maybe these discussions should happen on the many other places that want to attract and encourage them, and not so much on this little corner of the web that’s trying to be something different.
This is an article about a conservative tech company producing an AI that pushes their conservative talking points. The only common trait in these so called ‘low quality’ threads that the HN staff feel the _need_ to call out is that conservative look really bad. The tech is still just as interesting to discuss.
https://news.ycombinator.com/item?id=42992992
https://news.ycombinator.com/item?id=42996035
To me this is the most significant point:
What happens in flamewars is that when people encounter material they strongly disagree with, these systems get activated and rapidly produce aggressive and defensive responses that have to do with self-protection, and nothing to do with thoughtful consideration of the material, things one might learn, points where one might be wrong, curiosity, playful interaction, and so on. When survival is at stake there is no time or space for the latter sorts of reactions. But it's the latter that we want on HN—they're what the site is for.
I have my own experience over several years undertaking various forms of subconscious work, and from that experience have become very aware of the way emotional reactivity and sympathetic nervous system activation are antithetical to curious, reflective exploration of topics.
The evidence can be seen right here in this thread: how many insightful, reflective, curious comments are there in the entire thread? How many commenters are even attempting to comment in that style, or favour that style of cognitive processing of the topic?
What is wrong is to repeatedly use this website for a purpose other than its stated purpose and in breach of the guidelines, then when engaging in discussion with the moderators about the site's purpose and moderation approach, poison the discussion with these kinds of accusations and barbs.
The topic of biases and agendas has been raised countless times here over the years [1]. We're routinely accused of being biased in favour of one side or the other, or when that won't stick, of being "status-quo-ist" or some similar kind of “centrism” neg, none of which we can defend ourselves against without fuelling someone else's accusation of some other form of bias.
All we can do is keep explaining and demonstrating that the one thing we're trying to optimise for is intellectual curiosity. And in a world where it seems every major media outlet, social media platform and political actor trades on getting everyone riled up every day, we think it's important that there can be one corner of the web that isn't all about getting people riled up all the time.
The Wiggum insight: "My cat's breath smells like cat food."
It's probably best to give up with the website and use an RSS reader. I also turn on hidden comments. The censorship can be mostly avoided.
> People comment on flagging only to be ignored by the management
This still PG's site and I have no illusions about where his allegiances lie. The whole lot of them care for nothing but their own power and wealth.
Given his politics and heritage, I would assume he thinks this claim is true. But the rogue actor hacked Grok to make it say the claim is false.
(Man, I'd blocked him years ago. I hadn't realised how bad it'd gotten.)
This is the sort of outcome you'd expect from someone who had high level access including the ability to inject prompts but lacked the time and attention to detail to validate the actual results. Elon certainly isn't the only person that could match this description, but frankly it's not unlike him and how he reacts to things on social media...
It's totally fine if AI spreads right-wing conspiracy theories and propaganda, that's just... what are they calling it now... " maximal truth-seeking."
If you think about it, it makes a lot sense. Our human to human communications are actually rather rudimentary, we can't transfer much information. Instead, we all create a model of others based on our own ideas and experiences and whatever we think others are doing it is based on our own ideas.
Nobody is surprised when it turns out gatherings of powerful people are nests of corruption and malevolence. Eg, if I talk about the "bone saw incident" it isn't ambiguous who I mean - but the major actors are still welcome in polite society. That is the quality of person we're dealing with in positions of power - slightly extreme example, but still acceptable by global standards.
I didn't predict that it'd be injecting ideology instead of brand names because that seemed way too lame, but that's my blindspot I guess.
That is very different than the foundational technology falling into the hands of foreign actors.
> Wouldn't you want an inquery into their security procedures?
No more than I want an inquiry into SpaceX or Tesla, since X and XAi are not the same company.
Context matters.
Given that Xitter is still a fairly widely-used social network platform, and Grok is supposed to be a major part of its defense against misinformation, including misinformation about things like domestic elections and corruption, I would say that it very reasonably qualifies as such. (Granted, to a large extent that horse has left the barn—but that doesn't mean we should just burn down the barn.)
And to me that's a completely different concern than trying to limit another country's access to hardware for training large models, for example.
[0]https://www.reuters.com/technology/artificial-intelligence/m...
Sigh. Get PST/PDT right or just say PT.
(No, I don't think they were intending to speak from the perspective of Arizona or Hawaii, the only parts of the US that use PST but do not observe DST.)
Arizona is on year-round MST, Hawaii on year-round HST. Also, American Samoa is year-round SST, Guam and the Northern Marianas are in year-round CHST, and Puerto Rico and the US Virgin Islands are in year-round AST. (If you're keeping track, yes, all of the US territories do not use DST.) Indiana used to largely observe year-round EST, but the governor changed that several years ago, much to the chagrin of some of the Indianans I know.
The western-most parts of the Washington county are further west than the easternmost parts of some of the counties that are in America/Chicago.
But the center of commerce, information, entertainment, and all other infrastructure for Washington County, Indiana is Louisville, KY (lollvole). Louisville is America/New York.
When you make an appointment at UofL Jewish you don't care when the sun rises and sets or when the sun is directly overhead-- you care that the clock on your kitchen wall matches the clock in the doctor's office.
Washington County used to be as remote and desolate as you can think of but nowadays it is practically a suburb of Louisville. Farms along 150 are being turned into subdivisions as quickly as farmers can die off so their kids can offload the land.
Same thing with the north west part of the state. Chicago suburbs want to be synced up with Chicago. The southwest part of the state, specifically grain elevator operators and other businesses, want to be synced up with the logistics hubs of St. Louis and Nashville.
Most of the rest of the state is farms. They don't care and just do what Indianapolis does.
I've never understood people who say we have the jacked-up time zones because of farmers. The number of farmers who look at the clock before starting work for the day can be counted on zero fingers.
I grew up with "fast time" and "slow time" and once you are accustomed to it, it takes about as much mental effort as blinking.
Internet people want things to be orderly. Consistent. Algorithmic.
REAL people want to make it to their appointments on time and want the only business serving the area for 150 miles to be open when they call.
The IANA time zones use the definition of "has shared the same clocks since January 1, 1970," which means if a county in Indiana has switched from Eastern to Central (or vice versa) since that point, it gets a new time zone. The Eastern Time Zone portion of Indiana has switched from not observing DST to observing DST, which means it's separate from America/New_York.
Per the Wikipedia article, there's 11 IANA time zones in Indiana alone.
Except for the Navajo Nation, which spans parts of Arizona, New Mexico and Utah, and observes daylight saving time.
Which of course is the same mistake as the PDT/PST thing that sparked this whole thread.
But as to the more general sense that there's presumably some single global time, I actually expect that's common knowledge, you don't need to have thought about the details and realised how complicated it would be for this idea to be attractive. If anything I expect most people misunderstand from the wrong direction - they'd be surprised there isn't universal time because the general population grasp of relativity is too weak for them to see why it would be nonsense.
I would guess lots of the world population thinks of this single global time as GMT, the predecessor of UTC, but that's only like people thinking of SSL when they actually mean TLS, no big deal.
It surely isn't, because, well, everyone knows the time is different around the world and hence there's no global time.
If there really wasn't global time, this couldn't work, time zones, all of that jazz won't work if you have a system where relativity matters, for example enormous variation in height above sea level, a particularly enormous planet or very fast spin. Or if your perceptual scales are very different, but for us, here, there is global time.
https://greasyfork.org/en/scripts/534728-hacker-news-content...