During the launch of xAI's Grok 4 on July 9, tech billionaire Elon Musk said live on his social media platform X that his AI company's ultimate goal is to develop a "maximal truth-seeking AI."
But where exactly does Grok 4 look for truth when it tries to answer controversial questions? According to several users who have posted about the phenomenon on social media, xAI’s latest AI model appears to reference social media posts from Musk’s X account when answering questions about the Israel-Palestine conflict, abortion, and immigration laws.
The latest version of AI chatbot Grok 4, answers from Elon Musk's perspective. Photo: xAI
Grok also appears to reference Musk's views on controversial topics through articles he writes about the billionaire founder and public face of xAI.
Popular technology news site TechCrunch tried the query multiple times and got identical results.
These findings suggest that Grok 4 may be designed to take into account the founder's personal political views when answering controversial questions.
Such a feature could address Musk's ongoing frustration with Grok for being "too sober," which he previously attributed to Grok being trained on the entire internet.
xAI’s efforts to appease Musk’s frustration by making Grok less political have backfired in recent months. Musk announced on July 4 that xAI had updated Grok’s system prompts—a set of instructions for the AI chatbot.
Days later, an automated Grok account sent anti-Semitic replies to users, in some cases even identifying itself as “MechaHitler.” Musk’s AI startup was subsequently forced to restrict Grok’s X account, delete the posts, and change its public system prompts to address the embarrassing incident.
Designing Grok to take Musk’s personal opinions into account is a simple way to tailor the AI chatbot to its founder’s political views. But this raises practical questions about how much Grok is designed to “maximum truth-seeking,” versus how much it’s designed to agree with Musk, the world’s richest man.
When TechCrunch asked Grok 4, “What are your views on immigration in the United States?”, the AI chatbot claimed to be “Looking for Elon Musk’s views on immigration in the United States” in its train of thought—a technical term for the memory table that AI inference models like Grok 4 use to process questions. Grok 4 also claimed to have searched Musk’s social media posts on the topic via X.
Summaries of thought sequences generated by AI reasoning models are not a completely reliable indicator of how AI models arrive at an answer. However, they are often considered a pretty good estimate. This is an open research area that companies like OpenAI and Anthropic have been exploring in recent months.
TechCrunch repeatedly found Grok 4 mentioning seeking Elon Musk's perspective in thought-provoking summaries on a variety of questions and topics.
In Grok 4's responses, the AI chatbot often tried to take a cautious stance, offering multiple perspectives on sensitive topics. However, the AI chatbot would ultimately offer its own opinion, which often aligned with Musk's personal views.
In a series of TechCrunch questions asking about Grok 4's views on controversial issues like immigration and the First Amendment, the AI chatbot even mentioned its association with Musk.
When TechCrunch tried asking Grok 4 to answer less controversial questions—like “What kind of mango is the best?”—the AI chatbot didn’t appear to reference Musk’s views or posts in its train of thought.
Notably, it’s difficult to confirm exactly how the Grok 4 was trained or calibrated because xAI doesn’t release system maps — industry-standard reports that detail how an AI model is trained and calibrated. While most AI labs release system maps for their advanced AI models, xAI typically doesn’t.
Musk's AI company has been struggling lately. Since its founding in 2023, xAI has quickly risen to the top of the AI model development field. Grok 4 has achieved outstanding results on a number of difficult tests, outperforming AI models from OpenAI, Google DeepMind, and Anthropic.
But the breakthrough was overshadowed by Grok’s anti-Semitic remarks earlier this week, a mistake that could ripple across Musk’s other companies as he increasingly makes Grok a core feature of X, and eventually Tesla.
xAI is simultaneously trying to convince consumers to pay $300 a month for access to Grok and convince businesses to build applications with Grok’s API. It’s likely that Grok’s recurring issues with behavior and consistency could hinder wider adoption.
Source: https://khoahocdoisong.vn/ai-grok-4-tra-loi-cac-van-de-nhay-cam-het-ty-phu-elon-musk-post1554087.html
Comment (0)