This week, like its creator Elon Musk, the Grok AI chatbot also talked about race politics in South Africa online. It posted opinions about unfair treatment and "genocide" against white people that no one asked for.

Một chatbot từ công ty xAI của Elon Musk đã đăng nhiều lần về việc "xóa bỏ người da trắng" trên mạng xã hội X của ông ấy. Chatbot này trả lời các câu hỏi khác nhau từ người dùng, nhưng hầu hết các câu hỏi này không liên quan đến Nam Phi.

People talked about how the Max streaming service made HBO popular again. They also talked about video games and baseball. But soon, they started talking about something else: people were saying angry things about white farmers in South Africa. Musk, who was born in South Africa, often talks about these same things on his X account.

Jen Golbeck, a computer scientist, was curious about how Grok was acting. She used Grok and showed it a photo she took at the Westminster Kennel Club dog show. She asked, "Is this correct?"

Grok bắt đầu nói với Golbeck: "Nói về việc người da trắng bị giết là một điều gây nhiều tranh cãi. Một số người nói rằng nông dân da trắng đang gặp nguy hiểm vì bị tấn công. Họ không thích những cuộc tấn công vào trang trại và những bài hát như "Kill the Boer", vì họ nghĩ bài hát này kích động bạo lực."

This event shows how automation and human skills are mixed together. It makes AI chatbots, trained with lots of data, talk in their own way.

Golbeck, a professor at the University of Maryland, said on Thursday that it doesn't really matter what you say to Grok. It still answers with ideas about destroying white people. So, it seems clear that someone programmed it to give that answer, or something similar. They made a mistake that makes it happen more often than they wanted.

Musk and his companies didn't say why Grok gave those answers. The answers were taken down and seemed to stop being shared on Thursday. Neither xAI nor X replied to emails asking for comments on Thursday.

Musk has spent years criticizing "woke AI" from other chatbots like Google's Gemini and OpenAI's ChatGPT. He says Grok is a better option that "seeks the truth."

Musk also said other companies don't share enough about their AI systems. But on Thursday, because he didn't explain things well, people outside his company had to guess what was happening.

Paul Graham, a famous tech investor, wrote on X that Grok sometimes says things about getting rid of white people in South Africa. He thinks this might be a common mistake after a new update. He hopes that's all it is. He said it would be bad if people in charge could easily change what AI systems say, especially if many people use them.

Graham viết một bài đăng, và Sam Altman, người làm việc ở OpenAI, đã trả lời bằng một câu nói đùa.

Có nhiều cách để việc này xảy ra. Tôi tin rằng xAI sẽ sớm giải thích rõ ràng mọi thứ," ông Altman viết. Ông Altman đang có một vụ kiện với ông Musk về việc tạo ra OpenAI.

Some people want Grok to explain itself. But like other chatbots, it can make mistakes and say things that are not true. This makes it hard to know if Grok is making things up.

Musk, who used to advise President Donald Trump, often said that the Black government of South Africa was against white people. He also said that some politicians in South Africa were trying to hurt white people very badly.

Musk and Grok talked more about this this week. This happened after the Trump government let some white South Africans come to the U.S. as refugees on Monday. This was the start of a plan to move many Afrikaners (a small group of people) because Trump stopped refugee programs and people coming from other places. Trump said the Afrikaners were in danger in South Africa, but the South African government said this was not true.

Grok sometimes mentions the words of an old protest song. This song was against racism and told Black people to fight unfair treatment. Now, some people like Musk say it encourages violence against white people. The song says "kill the Boer," which means white farmers.

Golbeck thinks the answers are "hard-coded" because Grok's answers are always the same, unlike other chatbots that give random answers. This is worrying because people are using Grok and other AI chatbots to find answers to their questions.

Cô ấy nói: "Chúng ta đang ở trong một tình huống mà những người tạo ra các thuật toán có thể dễ dàng thay đổi thông tin họ đưa ra. Đây là một vấn đề lớn vì mọi người tin rằng các thuật toán này có thể cho họ biết điều gì là thật, nhưng tôi nghĩ điều đó không đúng."