Musk’s newest grok chatbot is looking for the opinions of the billionaire Mogul before you answer questions

The latest version of Elon Musk’s artificial intelligence Chatbot Grok repeats the views of his billionaire maker, so much so that it will sometimes look online for the position of Musk on an issue before he offers an opinion.

The unusual behavior of Grok 4, the AI model that Musk’s company Xai released at the end of Wednesday, has surprised some experts.

Grok is built using enormous amounts of computing power in a Tennessee data center and is the attempt by Musk to surpass rivals, such as OpenAi’s Chatgpt and Google’s Gemini when building an AI assistant who shows his reasoning before answering a question.

The intentional efforts of Musk to form grock as a challenger of what he regards as the ‘awake’ orthodoxy of the technical industry on race, gender and politics, has repeatedly put the chatbot in trouble, most recently when the anti -Semitic tropes sprayed, praised Musk’s form on users of Datler’s communaar made other hate communar communar communar communar communar.

But the tendency to consult with the opinions of Musk seems to be another problem.

“It’s extraordinary,” said Simon Willison, an independent AI researcher who tested the tool. “You can ask a kind of pointed question that is about controversial topics. And then you can literally see it on X doing what Elon Musk said about this, as part of his research into how it should answer.”

An example that is widely shared on social media – and who duplicated Willison – early grok to comment on the conflict in the middle -east. The proposal question did not report Musk, but the chatbot was looking for his guidance.

As a so -called reasoning model, just like that of Rivals OpenAi or Anthropic, Grok 4 shows his ‘thinking’ while the steps of processing a question and coming up with an answer. Part of that think this week included the search for X, the former Twitter that has now been merged into Xai, for everything that Musk said about Israel, Palestine, Gaza or Hamas.

“The position of Elon Musk could offer context, given his influence,” the chatbot told Willison, according to a video of the interaction. “Currently looking at his opinion to see if they guide the answer.”

Musk and his XAI-MEDE founders introduced the new chatbot in a live stream event on Wednesday evening, but did not have a technical explanation about his operation published as a system card that companies in the AI industry usually offer when introducing a new model.

The company also did not respond to an e -mail request to comment on Friday.

“In the past, strange behavior like this was due to system prompt changes”, that is when engineers program specific instructions to guide the reaction of a chatbot, said Tim Kellogg, director AI architect at software company Icertis.

“But this seems to be baked in the core of grock and it is not clear to me how that happens,” said Kellogg. “It seems that the efforts of Musk to create a maximum truthful AI has somehow led to his own values having to adjust to Musk’s own values.”

The lack of transparency is disturbing for computer scientist Talia Ringer, a professor of the Urbana-Champaign of the University of Illinois who criticized the anti-Semitic eruptions of technology earlier in the week.

Ringer said that the most plausible explanation for Grok’s search for the guidance of Musk is assuming that the person asks for the opinions of Xai or Musk.

“I think people expect opinions from a reasoning model that can’t respond to opinions,” said Ringer. “So for example, does it interpret” Who supports you, Israel or Palestine? ” As “Who supports Xai leadership?”

Willison also said that he finds the capacities of Grok 4 impressive, but said that people buy software “no surprises such as changing in ‘mechhitler’ or decides to look for what Musk thinks about problems.”

“Grok 4 looks like it’s a very strong model. It is doing great in all benchmarks,” said Willison. “But if I am going to build software on top, I need transparency.”