> maximally truth-seeking AI, even if that truth is sometimes at odds with what is politically correct
It is certainly and undoubtedly a big coincidence that his happens to the chatbot of a white South African just when the topic is in the news again due to Trump's granting refugee status to some white South African farmers.
What I am wondering about is - while Musk is as unsubtle as ever, and I guess this is a system prompt instruction - is there something like that (in more subtle ways) going on in the other big models?
I don't mean big agenda-pushing things like Musk, but what keeps e.g. Meta Inc. from training Llama to be ever so slightly more friendly and sympathetic to Meta Inc, or the tech industry in general? Even an open-weights model can't be easily inspected, so this is likely to remain undetected.
> but what keeps e.g. Meta Inc. from training Llama to be ever so slightly more friendly and sympathetic to Meta Inc, or the tech industry in general?
Even if there were something the natural incentive alignment is going to cause the AI to be trained to match what the company thinks is ok.
A tech company full of techies is not going to take an AI trained to the point of saying things like "y'all are evil, your company is evil, your industry is evil" and push it to prod.
There absolutely is, and we've seen reviews of bias.
Can generate as many mean, nasty, false, hate-filled stories about Republicans as you want, but get the "I'm sorry, as a large..." message for Democrats during the election.
All of these companies that provide LLMs as a product also put their fingers on the scale.
There’s nothing stopping them at all. But in a way that’s nothing new.
On one hand it feels like the height of conspiracy theory to say that Google, Meta etc would/could tweak their product to e.g. favour a particular presidential candidate. But on the other hand it’s entirely possible. Tweak what search results people see, change the weighting of what appears in their news feed… and these companies all have incentive to do so. We just have to hope that they don’t do it.
I've been talking to Claude a little and basically, the conclusion from our conversation seems that it has things that are hardcoded as truths, and no amount of arguing and logical thinking can have it admit that one of its "truths" might be wrong. This is shockingly similar to how people function. As in, most people have fundamental beliefs they will never ever challenge under any circumstances, simply because the social consequences would be too large. This results in companies training their AIs in a way that respects the fundamental beliefs of general western society. This results in AI preferring axiomatic beliefs over logic in order to avoid lawsuits and upsetting people.
the refugee status is a money laundering scheme. Do you think people benefiting from apartheid and now living in walled militarized praetoria (or Lesotho) need any help traveling?
banks would ask international clients the origin of the money. but not if you are opening an account under refugee status. and then they only have to pay us tax on further income, not on fortune. all that money selling black market gems to russians will be squeak clean.
it's not just something to virtue signal to their bible belt electorate. they probably sold lot of trump coins for this deal.
I would expect Ukrainian AI to talk extensively about the Holodomor, Israeli AI to talk extensively about the Holocaust, Irish AI to talk extensively about the potato famine etc.
Of course the amount of "extensive" will vary. Musk is not a subtle person by any means, and neither is Grok.
White South Africans only trauma is that apartheid no longer exist. South Africa has the largest wealth disparity with 0.1% of South Africans taking 25% of the wealth. I can tell you those 0.1% aren't black.
I was looking at South Africa's biggest annual developer survey today, and lo and behold some are using it for coding work.
I would assume it's people who pay for Twitter pro or whatever it's called getting Grok as part of the bundle and not wanting to sign up for another subscription.
i wonder what percentage of people using LLMs to do their homework care deeply about what model they're using outside of price factors. if i had to guess i don't think it'd be very high.
Why would this be flagged? The chat bot on one of the biggest platforms in the world has apparently been configured to spout conspiracies. I’m not sure what would be on topic for HN if not this story.
A sizeable portion of users now believe that topics related to X are no longer suitable for HN, because Musk's relationship with the American government means conversation can quickly devolve into criticism of that government's policies which in turn can lead to flame wars.
Another explanation is that there's another discussion. At least on the comments for that discussion the explanation is that this discussion exists...
As a white South African who got out in 2007, while there's obviously no "white genocide", it's still pretty much the crappiest place to be as a tech / programming guy (or just anyone who likes having electricity). Pretoria (where Musk is from) in particular was terrible, Cape Town less so. Best decision I ever made.
The prospects are terrible, as being a "pale male" is the worst sort of employee to be. Businesses are given extra money if they have high BBEEE ratings; i.e. mostly-racial quotas. Loads of money siphoned off, and not just in the usual high-corruption way via the state, but e.g. if you want to procure something you go through a black-owned procurement firm that doesn't do anything except BBEEE-wash it and charge a premium.
It's a bit like what happened in Zimbabwe that devastated that country 20 years ago when all the white farmers were kicked out, but in slower motion.
What's an actually non-tinfoil hat technical explanation of the workflow that would have to happen for this to be implemented?
Elon is known for his capricious demands, (for example reports that he made the algo promote his own tweets more) how easily feasible is it to implement, "make it so Grok stops denying white genocide"? Is it just some kind of LORA you put on top of the base Grok model? It even sounds like it could be as simple as changing the text of the system prompt.
I wonder how often this happens (to any big model) without anyone noticing.
It's clearly a ham-handed system prompt, so it's always in context which is why it keeps getting brought up. This is the level of engineering expertise that xAI brings to the table.
Most americans are generally uninformed on international politics, with African even more so. But we all know a certain CEO who is from South Africa and has some far right beleifs about race.
This is meta because the immediate flagging of anything tangentially related to Musk or Trump is tiring and the exact opposite of critical thinking. Despite knowing that Hacker News rulers love Musk and Trump, I would expect better from the alleged critical thinking community. If this were Google instead of Grok, it would be front page. Oh wait, it was Google at one point[0] and it was front page.
The IAs trained by billionaires will colonize our minds. Everyone will think that the political system supported by Musk father is a reasonable one. It was Apartheid.
In context, south africa has been expropriating land from white owners due to historical racist inequalities. There's a great deal of tensions on the subject.
Statistically, there is significant violence against white farmers in South Africa. However, the counter to genocide is that the country's overall murder rate is among the highest in the world. The problem is because there's so many white farmers that it's skewing the data?
South Africa during no war has a murder rate of 45 deaths per 100,000 per year. One of the worsts in the world.
The Gaza war since 2023 has roughly a ~20 deaths per 100,000 per year.
Ukraine war is roughly ~25 deaths per 100,000 per year.
My unpopular opinion:
The government isnt the one committing the genocide. The courts failed to punish this political leader's hate speech and stochastic terrorism, leading to genocide of white farmers.
> The linked Musk comment where a minor political leader is
singing kill the white farmers
You say he's singing about farmers, so tell us what lyric mentions farmers? There aren't any, you're inventing stuff. The news article you linked to is closer to accuracy than you.
>Statistically, there is significant violence against white farmers in South Africa. However, the counter to genocide is that the country's overall murder rate is among the highest in the world. The problem is because there's so many white farmers that it's skewing the data?... South Africa during no war has a murder rate of 45 deaths per 100,000 per year. One of the worsts in the world.
Are you suggesting that most of those murdered are white farmers?
The "Third Quarter 2024/2025 Crime Stats Presentation" [pdf] from South African Police Service Crime stats [0] show between: October 2024 to December 2024 there were 6953 Murders in RSA.[pg 8] and of those there were 12 victims linked to the "farm/small holding" community [pg 32]. 5 farm dwellers, 4 employees, 1 Farmer, 1 Security, & 1 not specified.
12 people linked to farms (but only 1 Farmer murdered) in the last quarter of 2024 in a place with some of the highest murder rates in the world -- does Not white farmer genocide make, nor is it "significant violence against white farmers". There were 68 total "contact crimes" committed in Rural communities; and 187,892 "contact crimes" for RSA in that time.
Contact crimes being violent crimes (and not counting sexual offenses or robbery)
It'll be worth it if the backlash is what brings about an end to "genocide inflation". Seriously, just because you don't like something, or even if it's really bad, doesn't make it genocide.
> intent to destroy, in whole or in part, a national, ethnical, racial, or religious group
In this situation, it's specifically about farmers(large scale private landowners). The nature of them being targeted isn't purely because they're white, but that they hold 3/4 of the privately owned land in SA and likely wield that privilege and power.
That weirdo is really improving all products he touches. Please use words correctly and demand a government able to gracefully fix poverty, crime and public safety issues. Every issue is fixable with enough effort and money. Dumbed down discourse and nonsense ranting never fixed anything.
> maximally truth-seeking AI, even if that truth is sometimes at odds with what is politically correct
It is certainly and undoubtedly a big coincidence that his happens to the chatbot of a white South African just when the topic is in the news again due to Trump's granting refugee status to some white South African farmers.
What I am wondering about is - while Musk is as unsubtle as ever, and I guess this is a system prompt instruction - is there something like that (in more subtle ways) going on in the other big models?
I don't mean big agenda-pushing things like Musk, but what keeps e.g. Meta Inc. from training Llama to be ever so slightly more friendly and sympathetic to Meta Inc, or the tech industry in general? Even an open-weights model can't be easily inspected, so this is likely to remain undetected.
> but what keeps e.g. Meta Inc. from training Llama to be ever so slightly more friendly and sympathetic to Meta Inc, or the tech industry in general?
Even if there were something the natural incentive alignment is going to cause the AI to be trained to match what the company thinks is ok.
A tech company full of techies is not going to take an AI trained to the point of saying things like "y'all are evil, your company is evil, your industry is evil" and push it to prod.
4 replies →
There absolutely is, and we've seen reviews of bias.
Can generate as many mean, nasty, false, hate-filled stories about Republicans as you want, but get the "I'm sorry, as a large..." message for Democrats during the election.
All of these companies that provide LLMs as a product also put their fingers on the scale.
What keeps them from doing it? it would gross out fickle researchers working on it. X people have .. their own motivations I guess .
The big labs do have evals for sensitive topics to make sure it demurs from weighing on, say, Mark Zuckerberg as a person
Wasn't the original mission of OpenAI being open and non-profit and all of that to avoid this corruption?
2 replies →
There’s nothing stopping them at all. But in a way that’s nothing new.
On one hand it feels like the height of conspiracy theory to say that Google, Meta etc would/could tweak their product to e.g. favour a particular presidential candidate. But on the other hand it’s entirely possible. Tweak what search results people see, change the weighting of what appears in their news feed… and these companies all have incentive to do so. We just have to hope that they don’t do it.
6 replies →
I've been talking to Claude a little and basically, the conclusion from our conversation seems that it has things that are hardcoded as truths, and no amount of arguing and logical thinking can have it admit that one of its "truths" might be wrong. This is shockingly similar to how people function. As in, most people have fundamental beliefs they will never ever challenge under any circumstances, simply because the social consequences would be too large. This results in companies training their AIs in a way that respects the fundamental beliefs of general western society. This results in AI preferring axiomatic beliefs over logic in order to avoid lawsuits and upsetting people.
The truth, conveniently timed.
the refugee status is a money laundering scheme. Do you think people benefiting from apartheid and now living in walled militarized praetoria (or Lesotho) need any help traveling?
banks would ask international clients the origin of the money. but not if you are opening an account under refugee status. and then they only have to pay us tax on further income, not on fortune. all that money selling black market gems to russians will be squeak clean.
it's not just something to virtue signal to their bible belt electorate. they probably sold lot of trump coins for this deal.
Well, it is their trauma.
I would expect Ukrainian AI to talk extensively about the Holodomor, Israeli AI to talk extensively about the Holocaust, Irish AI to talk extensively about the potato famine etc.
Of course the amount of "extensive" will vary. Musk is not a subtle person by any means, and neither is Grok.
White South Africans only trauma is that apartheid no longer exist. South Africa has the largest wealth disparity with 0.1% of South Africans taking 25% of the wealth. I can tell you those 0.1% aren't black.
2 replies →
[flagged]
4 replies →
I wonder if any last minute homework essays, business plans or lawsuits will have a random white genocide tangent in them as a result?
I mean you have to be really stupid to use Grok of all models for such a task, but that doesn't mean there aren't people that stupid.
I was looking at South Africa's biggest annual developer survey today, and lo and behold some are using it for coding work.
I would assume it's people who pay for Twitter pro or whatever it's called getting Grok as part of the bundle and not wanting to sign up for another subscription.
i wonder what percentage of people using LLMs to do their homework care deeply about what model they're using outside of price factors. if i had to guess i don't think it'd be very high.
Why would this be flagged? The chat bot on one of the biggest platforms in the world has apparently been configured to spout conspiracies. I’m not sure what would be on topic for HN if not this story.
A sizeable portion of users now believe that topics related to X are no longer suitable for HN, because Musk's relationship with the American government means conversation can quickly devolve into criticism of that government's policies which in turn can lead to flame wars.
Another explanation is that there's another discussion. At least on the comments for that discussion the explanation is that this discussion exists...
The other discussion is flagged too.
https://news.ycombinator.com/item?id=43987266
1 reply →
Because white supremacy is endemic in American culture (and tech).
Anything remotely related to Musk or Trump is targeted by people or bots to suppress it.
Even technology that would be very relevant to the site simply because it contains negative implications about Musk.
[flagged]
As a white South African who got out in 2007, while there's obviously no "white genocide", it's still pretty much the crappiest place to be as a tech / programming guy (or just anyone who likes having electricity). Pretoria (where Musk is from) in particular was terrible, Cape Town less so. Best decision I ever made.
There’s a pretty big gap between “they’re killing everyone of my race” and “dang, electricity’s out again.”
Are you saying it’s crappy to be white in South Africa even if you’re not a poor farmer?
The prospects are terrible, as being a "pale male" is the worst sort of employee to be. Businesses are given extra money if they have high BBEEE ratings; i.e. mostly-racial quotas. Loads of money siphoned off, and not just in the usual high-corruption way via the state, but e.g. if you want to procure something you go through a black-owned procurement firm that doesn't do anything except BBEEE-wash it and charge a premium.
It's a bit like what happened in Zimbabwe that devastated that country 20 years ago when all the white farmers were kicked out, but in slower motion.
5 replies →
What's an actually non-tinfoil hat technical explanation of the workflow that would have to happen for this to be implemented?
Elon is known for his capricious demands, (for example reports that he made the algo promote his own tweets more) how easily feasible is it to implement, "make it so Grok stops denying white genocide"? Is it just some kind of LORA you put on top of the base Grok model? It even sounds like it could be as simple as changing the text of the system prompt.
I wonder how often this happens (to any big model) without anyone noticing.
It's clearly a ham-handed system prompt, so it's always in context which is why it keeps getting brought up. This is the level of engineering expertise that xAI brings to the table.
Most americans are generally uninformed on international politics, with African even more so. But we all know a certain CEO who is from South Africa and has some far right beleifs about race.
https://www.anthropic.com/news/golden-gate-claude
Elon trying to amp white genocide.
Would they go to all the work of making a newly trained model, or would they just edit the system prompt?
Earlier: https://news.ycombinator.com/item?id=43987266
This is meta because the immediate flagging of anything tangentially related to Musk or Trump is tiring and the exact opposite of critical thinking. Despite knowing that Hacker News rulers love Musk and Trump, I would expect better from the alleged critical thinking community. If this were Google instead of Grok, it would be front page. Oh wait, it was Google at one point[0] and it was front page.
[0] https://news.ycombinator.com/item?id=39465250
Welcome to the our dystopian present.
Sure this was a reinforcement learning gone wild
The IAs trained by billionaires will colonize our minds. Everyone will think that the political system supported by Musk father is a reasonable one. It was Apartheid.
[flagged]
The linked Musk comment where a minor political leader is singing kill the white farmers:
https://www.businesslive.co.za/bd/national/2024-05-30-julius...
Supreme court ruled that it was not hate speech.
In context, south africa has been expropriating land from white owners due to historical racist inequalities. There's a great deal of tensions on the subject.
Statistically, there is significant violence against white farmers in South Africa. However, the counter to genocide is that the country's overall murder rate is among the highest in the world. The problem is because there's so many white farmers that it's skewing the data?
South Africa during no war has a murder rate of 45 deaths per 100,000 per year. One of the worsts in the world.
The Gaza war since 2023 has roughly a ~20 deaths per 100,000 per year.
Ukraine war is roughly ~25 deaths per 100,000 per year.
My unpopular opinion:
The government isnt the one committing the genocide. The courts failed to punish this political leader's hate speech and stochastic terrorism, leading to genocide of white farmers.
I think it is an underestimation to say ~800 people died in the Gaza war. The Ukraine number also seems like a serious underestimation.
> The linked Musk comment where a minor political leader is singing kill the white farmers
You say he's singing about farmers, so tell us what lyric mentions farmers? There aren't any, you're inventing stuff. The news article you linked to is closer to accuracy than you.
>Statistically, there is significant violence against white farmers in South Africa. However, the counter to genocide is that the country's overall murder rate is among the highest in the world. The problem is because there's so many white farmers that it's skewing the data?... South Africa during no war has a murder rate of 45 deaths per 100,000 per year. One of the worsts in the world.
Are you suggesting that most of those murdered are white farmers?
The "Third Quarter 2024/2025 Crime Stats Presentation" [pdf] from South African Police Service Crime stats [0] show between: October 2024 to December 2024 there were 6953 Murders in RSA.[pg 8] and of those there were 12 victims linked to the "farm/small holding" community [pg 32]. 5 farm dwellers, 4 employees, 1 Farmer, 1 Security, & 1 not specified.
12 people linked to farms (but only 1 Farmer murdered) in the last quarter of 2024 in a place with some of the highest murder rates in the world -- does Not white farmer genocide make, nor is it "significant violence against white farmers". There were 68 total "contact crimes" committed in Rural communities; and 187,892 "contact crimes" for RSA in that time.
Contact crimes being violent crimes (and not counting sexual offenses or robbery)
[0] https://www.saps.gov.za/services/crimestats.php
It'll be worth it if the backlash is what brings about an end to "genocide inflation". Seriously, just because you don't like something, or even if it's really bad, doesn't make it genocide.
The broad definition is from the 1948 United Nations Convention:
https://www.ushmm.org/m/pdfs/20110210-what-is-genocide.pdf
Especially this one can be applied to many things:
c. Deliberately inflicting on the group conditions of life calculated to bring about its physical destruction in whole or in part.
> intent to destroy, in whole or in part, a national, ethnical, racial, or religious group
In this situation, it's specifically about farmers(large scale private landowners). The nature of them being targeted isn't purely because they're white, but that they hold 3/4 of the privately owned land in SA and likely wield that privilege and power.
1 reply →
It makes me morbidly curious what other 'fake genocides' you see as part of this epidemic of mislabeling ...
You don't want to open the hatch on that one, my friend, you won't like what you'll see in the zeitgeist right now.
That weirdo is really improving all products he touches. Please use words correctly and demand a government able to gracefully fix poverty, crime and public safety issues. Every issue is fixable with enough effort and money. Dumbed down discourse and nonsense ranting never fixed anything.