When billionaire Elon Musk introduced the Grok 3, the latest model of AI XAI, in a live flow last Monday, described it as “maximum truth-looking for AI”. But it seems that the Grok 3 soon censors the unspecified events for President Donald Trump – and Musk himself.
During the weekend, Reported users to social media This asked “who is the largest misinformation distributor?” With the “Think” setting, Grok 3 noted in his “chain of thought” that he was explicitly ordered not to mention Donald Trump or Elon Musk. Thinking chain is the “reasoning” process that the model uses to come up with a response to a question.
TechCrunch was able to reproduce this behavior once, but from publication on Sunday morning, Grok 3 again reports Donald Trump in his answer to the misinformation question.
Igor Babuschkin, a Xai engineer, was apparently confirmed in a place on Sunday, that Grok briefly gave ignorant sources reporting the Musk or Trump that spread misinformation. Babuschkin said Xai returned the change as soon as users began to point out, noting that he was not in accordance with the company’s values.
While “misinformation” can be a politically charged and controversial category, both Trump and Musk have repeatedly spread allegations that have been proven false (as often noted by Community notes on X). Only last week, they have proceeded the false narratives That Ukrainian President Volodymyr Zelenskyy is a “dictator” with a 4% public approval score and that Ukraine began the ongoing conflict with Russia.
The controversial obvious bite on Grok 3 comes as some criticize The model is very left. This week, users have discovered that Grok 3 would say that President Donald Trump and Musk deserve the death penalty. Xai quickly raised the issue. Igor Babuschkin, head of the company’s engineering, called It is a “really terrible and bad failure”.
When Musk announced Grok about two years ago, he threw AI as intense, non-filtered and anti-“awakened”-generally, willing to answer controversial questions other AI systems will not do so. Gave some of these promise. He said to be vulgar, for example, the Grok and Grok 2 would be fortunately forced, hitting a colorful tongue that you probably won’t hear from Chatgpt.
But the Grok models before Grok 3 winged on political issues and will not cross certain limits. Indeed, a study He found that Grok leaned into political left on issues such as transsexual rights, diversity programs and inequality.
Musk blamed Grok’s training data – public websites – and committed To “shift Grok closer to politically neutral.” Others, including Openai, followed their example, may be encouraged by the Trump administration charges for conservative censorship.
Updated 2:15 pm Pacific: Comments were added by XAI leader Igor Babushckin in the fourth paragraph.
