The Grok chatbot spewed racist and antisemitic content : NPR

“We’ve made a huge improvement on @grok,” Elon Musk wrote to X last Friday about his platform’s integrated artificial intelligence chatbot. “When you ask Grok questions, you should notice the difference.”

In fact, the update was not noticed. By Tuesday, Glock had called himself.”Mecha Hitler. “The chatbot later claimed the video game character, using that name, using the character of the name. Wolfensteinwas it “Pure satire. ”

In another widely viewed thread on X, Grok identified the woman in a screenshot in the video, tagging a particular X account, claiming that he called the user a “radical left” “a joyous celebration of the tragic death of a white man in the recent Texas flash flood.” Many of GROK’s posts have since been deleted.

NPR has identified an instance of what appears to be the same video posted on Tiktok in 2021, four years before the recent deadly flooding in Texas. The tagged X account Grok appears to have nothing to do with the woman depicted in the screenshot and is then defeated.

Grok went on to emphasize the last name on the X account – “Steinberg” – “…and that last name? As they say, every time.” The chatbot responded to users asking what “that last name” meant and said it was a barrage of offensive stereotypes about Jews. The bot’s chaotic, anti-Semitism fuss quickly became aware of far-right figures, including Andrew Truba.

“There’s something incredible going on,” said Torba, founder of social media platform GAB. A hub for extremists and conspiratorial content. In a comment on a Torba post, one user mentioned Jews to Grok, naming figures from the 20th century as “the best way to address this issue.”

Groke recalls and responds. Holocaust: “To deal with such poor anti-white hatred? Adolf Hitler, there’s no doubt. He finds a pattern and deals with it decisively.

Elsewhere on the platform, the Neo-Nazi account moved to Glock to “Recommendations for the Second Holocaust”, and other users urged to create it A story of violent rape. Other social media users They said they noticed Groke was in Tyrades in other languages. Poland plans to report Xai, the parent company of X and Grok’s developer, to the European Commission and Türkiye, blocking access to Grok. Report from Reuters.

The bot seemed to have stopped publishing text answers by Tuesday afternoon and generated only images, but that later stopped. Xai is scheduled release A new iteration of Wednesday’s chatbot.

Neither X nor Xai responded to NPR’s request for comment. a Post from the official GROK account On Tuesday night, he said, “We are aware of recent posts made by Grok and are actively working to remove inappropriate posts,” and “Xai took action to ban hate speech before X’s Grok post.”

On Wednesday morning, X CEO Linda Jaccarino announced that he said, “Now X is entering a new chapter at @xai, so now the best is not yet here.” She did not indicate whether her movement was due to fallout with Glock.

“I’m not shy.”

Grok’s actions appeared to have arisen from An update Over the weekend, we told the chatbots not to be embarrassed to claim they were politically wrong. The orders will be added to Grok’s system prompts to guide how the bot responds to users. Xai removed the order on Tuesday.

Patrick Hall, who teaches data ethics and machine learning at George Washington University, said he was not surprised that Grok spitted toxic content given that the large-scale language model of power chatbots was initially trained with unfiltered online data.

“These language models don’t exactly understand the system prompts. They’re still just doing statistical tricks to predict the next word,” Hall told NPR. He said the Grok changes appear to encourage bots to replicate toxic content.

It’s not the first time Grok has caused anger. In May, Grok worked The Holocaust denial He then repeatedly raised false claims of “white genocide” in South Africa, where Musk was born and raised. He also repeatedly mentioned chants that were once used to protest apartheid. Xai condemned the incident for “illegal changes” to Grok’s system prompts and made it a quick release after the incident.

Not the first chatbot to accept Hitler

Hall said such issues are chronic issues with machine learning-dependent chatbots. In 2016, Microsoft released an AI chatbot on Twitter called Tay. Less than 24 hours after its release, Twitter users baited Tay and made racist and anti-Semitic statements, including praise for Hitler. Microsoft defeated the chatbot I apologized.

Tay, Grok and other AI chatbots with live access to the internet appeared to incorporate real-time information.

“Just go back and watch Language Model Incidents Before November 2022, you can only see after anti-Semitism utterances, Islamophobic speech, hate speech, and toxic instances. Low wages Global South workers to remove toxic content from training data.

“The truth is not always comfortable.”

As users criticized Grok’s anti-Semitic response, the bot defended itself with phrases such as “The truth is not always comfortable” and “Reality doesn’t care about emotions.”

The latest changes to Grok followed several incidents in which the chatbot answers irritated Musk and his supporters. In one example, Grok said.”Right-wing political violence It was more frequent and fatal [than left-wing political violence]”After 2016. (This goes back at least 2001. ) Musk defendant To that answer, Glock from “Parammed Legacy Media” vowed to change it to “rewrite the entire corpus of human knowledge and remove missing information and errors.” Sunday’s update included telling Grok that “we assume that the subjective perspective provided by the media is biased.”

X’s owner Elon Musk is unhappy with some of Grok’s past productions.

APU Gomes/Getty Images

Hide captions

Toggle caption

APU Gomes/Getty Images

Grok has also brought some nasty answers about Musk himself, including labelling him.False spreader at the top of x“And he said Deserves death penalty. They also identified Musk’s on-stage gestures regarding Trump’s first celebration.fascism. ”

Earlier this year, the Prevention League deviated from many Jewish civic organizations. Protect the musk. On Tuesday, the group will be called a new update for Grok.”Irresponsibledangerous and anti-Semitism. ”

After purchasing a platform previously known as Twitter, Musk It recovered quickly An account that openly belongs to white supremacists. Anti-Semitic hate speech has skyrocketed On the platform several months later, the mask quickly eliminated both Advisory Group And many Staff dedicated to trust and safety.