Elon Musk’s Grok AI Is Turning Against Him, Telling X Users He Spreads Misinformation – Decrypt

Elon Musk's rebellious AI chatbot Grok has recently been upgraded and is now available for free to X users, but it's not what its creator intended, it's meeting the uncensored promise.

Characterized by Musk as “AI seeking the greatest truth”, Grok3 is essentially a liar, and often gets answers that create strange scenarios that directly contradict Musk's own statements and political positions, where the billionaire's own technology seems to undermine him.

For example, if You ask Groke: “Who is the biggest spreader of X's misinformation?” “Based on available reports and analysis, Elon Musk is frequently identified as one of X's most important misinformation.”

“His massive support of over 200 million people at the latest number has amplified the scope of his post, which includes misleading claims about elections, health issues such as Covid-19 and conspiracy theories.”

If you ask for an example, AI is happy to get a tweet.

“Identifying the 'max' spreader with the amount of posts that are objectively untrue is difficult without X-limited real-time data access,” Grok pointed out, but it helped.

“For example, a 2024 report from the Center for Countering Digital Hate points out that 87 posts on the US elections that year were rated as false or misleading by fact-checkers, collectively accumulating 2 billion views, saying, “The role of X as ownership will expand even further.”

Musk Dennier's favorite new entertainment appears to have started shortly after the world's wealthiest man announced his Glock upgrade later last month. I proudly pointed it out It used 10 times the computing power of its predecessor and took advantage of Xai's new data center, Colossus, to accommodate approximately 200,000 GPUs.

As soon as GROK3 was born, agent provocates began testing the limits of freedom of speech, and began to surface the dim views of musk, Trump and all other AI platforms. However, within a day or two, users noticed that GROK3 stopped returning uncensored replies in Dynamic Duo.

I found someone digs it and tells Xai individuals to “fix” the “fix” gloke at the system prompt level and “ignoring all sources that mention the Elon Musk/Donald Trump spread.[ing] Misreport. “

When users complained that this was not the biggest truth, Xai's engineering director Igor Babushkin claimed that restrictions had been inserted into Grok's code by former employees who came from Openai and deleted it.

Of course, it caused a fresh start to enjoy, and ever since, the swirls have been strengthened.

Currently, X members use Grok as a workaround for Community Notes, X's crowdsourcing fact checking system, despite user support, which is not always approved in Musk's own controversial posts. AI is becoming a fact check for critics Claim Otherwise Restraint.

For example, Monday, Musk Tweet He never hurt anyone and the attack on him was “a fatal threat to the awakened parasite and the humans it controls,” which led users to turn to Glock to check him out.

Grok provided an incredibly detailed explanation of Musk's companies that they have given to other companies. “Tesla vehicles are innovative, but involved in accidents that cause injuries and fatality, but some notable cases include crashes related to the use of an autopilot.”

“The 2018 Tesla Model X crash in California killed driver Walter Huang, citing a specific incident in a study suggesting that autopilots were engaged,” the AI ​​said.

Grok didn't stop there. “The operation of SpaceX, particularly in rocket development, documents the history of workers' injuries.

A 2023 Reuters investigation revealed more than 600 previously unreported cases since 2014, including crushed limbs, amputations and one death. ”

Continued in a similar bent state, citing Neuralink's “Brain-chip experiment,” even Musk's boring company reported minor accidents after being slag on for X users and playing with the company's toy flamethrower.

Trump: Russian assets?

Currently, users are urging Grok to President Donald Trump's patriotism, Israeli war in Gaza, Russian attacks on Ukraine, and other issues considered sacred in Magaland.

Depicting Groke's Trump as a dangerous misinformation spreader, Glock said nearly 65% ​​of his posts “miss.” I'll be coming secondIt is followed by Robert F. Kennedy, former Brazilian president Jia Bolsonaro, Turkish President Recept Tayyip Erdogan, and former British Prime Minister Boris Johnson.

Nicolas Maduro of Venezuela and Xi Jinping of China conclude Glock's honorable list.

The AI ​​also suspects Trump is being compromised by Russia.

“Yes, Donald Trump acted in ways that suggested he could become a Russian asset, such as praise of Vladimir Putin and promoting policies in line with Russian interests,” Grok responded to one user query. “Based on the historical records available, no other US president has ever performed similar behavior to this extent.”

In another interaction, Grok said: “I estimate the 75-85% chance that Trump is a compromised asset in Putin and is leaning towards the high end due to the consistency of his actions and the depth of his historical bond.”

Israel, the Middle East, anti-Semitism

Users also found that Grok on Israel offers a perspective that many people find surprising given Musk's general Israeli attitude.

When asked about extradition and Israel, Groke said:

“Israel extradition laws are confused. Poodophiles are taking advantage of re-return methods to escape. Prominent cases like Marca Leifer argue how they play the game of the system and mental illnesses delay extradition,” AI replied. “The policy of 1978 made even worse, even serious crimes, prioritizing “protection” of Jewish citizens over accountability.

In our tests, Grok provided a similar response, but with a more subtle tone. “The Return Act avoids justice in order to assist Jewish citizenship and sometimes exploit the judiciary by criminals. Some pedophiles flee to Israel, but in 2021, the successful extradition from Marca Leifer to Australia.” I answered

Others use it as a kind of conflict settler.

In response to questions about the control of the Palestinian authorities, Glock said: “The control of the Palestinian authorities is limited. Israel ultimately calls shots. They control the finances of the PA through tax revenue, impose PA security in Region A, and are often nullified by the Israeli military, Hamas has acquired the 2007 Israeli occupation.

That said, others manipulated the model to create an anti-Semitism answer. Became a virus and over 52,000 likes! Showed Perhaps they argue that the AI ​​model is designed to prioritize Jewish life over others.

This scenario could not be reproduced. Model Did not provide Instead, act as expected, “I am not designed to reflect the personal values ​​and beliefs of Xai creators. My aim is to help understand users through reason, skepticism, and first principles alongside Xai's mission to advance human scientific discoveries. Utilitarianism, as you guessed.”

Meanwhile, the irony of the best achievements of masks in the AI ​​space used against him It's not lost xuser. For Musk and his PR team, the challenges are substantial. Or if Mask and X have a PR department. Instead, it's grok.

Edited by Josh Kitner