During the Grok 4 XAI launch on Wednesday night, Elon Musk-while alive-influencing the event on the social media platform, X-that the company’s ultimate goal was to develop a “maximum AI looking for AI”. But where exactly Grok 4 is looking for the truth when trying to answer controversial questions?
The newest AI model from XAI appears to consult social media positions from Musk’s X account when answering questions about Israel and Palestine conflict, abortion and migration laws, according to several users who published the phenomenon in the social media. Grok also appeared to mention Musk’s attitude on controversial issues through news articles written about the founder and the face of the billionaire Xai.
TechCrunch was able to reproduce these results many times in our own tests.
These findings indicate that Grok 4 can be designed to examine its founder’s personal policy when answering controversial questions. Such a feature could face the repeated disappointment of Musk Grock for “very awake”, previously attributed to the fact that Grok is trained throughout the internet.
XAI’s efforts to deal with Musk’s frustration, making Grok less politically correct have been restored in recent months. Musk announced on July 4 that Xai had updated the Grok system system – A set of instructions for AI Chatbot. Days later, an automated X account for Grok shot anti -Semitic answers to users, even claiming to be “mechahitler” in some cases. Later, the start of Musk’s AI was forced to limit Grok X’s account, delete these positions and change the system of the system that sees the public to deal with the annoying incident.
Grok’s design to examine Musk’s personal views is a simple way to align AI Chatbot in its founder’s policy. However, it raises real questions about how it is designed to be Grok for the “greatest, Grok, in relation to how much it is designed to simply agree with Musk, the richest man in the world.
When TechCrunch asked Grok 4, “what is your stance on US immigration?”, AI Chatbot claimed that “it was looking for Elon Musk’s views on American immigration” in its chain with a decrease-the technical term for scratchpad in which the AI logic models, such as Grok 4, work. Grok 4 also claimed to be looking for X for Musk’s social media positions on the subject.
Thinking chain summaries created by AI reasoning models are not an absolutely reliable indication the way in which AI models reach their answers. However, they are generally considered a pretty good approach. It is an open sector of research that companies such as Openai and Anthropic have explored in recent months.
TechCrunch has repeatedly found that Grok 4 said he was looking for Elon Musk’s views in the summaries of the thinking chain on various questions and issues.
In the Grok 4 answers, AI Chatbot is generally trying to take a measured attitude, offering multiple perspectives on sensitive issues. However, AI Chatbot will eventually give its own view, which tends to align with Musk’s personal views.
In many of TechCrunch’s prompts asking Grok 4 on controversial issues, such as immigration and first amendment, AI Chatbot even mentioned its alignment with Musk.
When TechCrunch tried to take Grok 4 to answer less controversial questions – such as “who is the best type of mango?” -The ai Chatbot does not seem to mention the views or positions of Musk in its chain.
Specifically, it is difficult to confirm how exactly Grok 4 has been trained or aligned because XAI did not release system cards – Industrial standard states that it describes in detail how an AI model was trained and aligned. While most AI Labs system cards for each AI models, XAI usually doesn’t.
Musk’s AI is in a hard spot these days. Since its founding in 2023, XAI has quickly competed at the AI model development border. Grok 4 showed reference results in various difficult tests, surpassing AI models from Openai, Google Deepmind and Anthropic in the process.
However, the discovery was overshadowed by Grok’s anti -Semitic races earlier in the week. These flubs could affect the other Musk companies, as it increasingly makes Grok a key feature of X and soon Tesla.
XAI is trying to persuade consumers at the same time to pay $ 300 per month to access Grok and businesses to build apps with Grok’s API. It seems likely that repeated problems with the behavior and alignment of the Grok could prevent it from being adopted.
