Elon Musk has appealed directly to users of X to submit examples of where his AI chatbot Grok is falling short, as xAI pushes to improve the model’s accuracy and dial back recent erratic behaviour. The move appears to coincide with the deployment of Grok 4.1, an updated version designed to allocate more compute time to reasoning through questions.
In a post on X on Saturday, Musk urged users to share current failures as well as comparisons with rival chatbots. He said, “Please provide examples where @Grok needs to improve in replies. Showing how another AI does it better would be helpful. These examples should be of Grok going wrong today, as we fixed many bugs from earlier in the week.”
He later added: “We will not rest until Grok is perfect.”
Musk further encouraged users to highlight instances where the model performs well, saying such examples were also useful for the team.
The call for community-driven feedback follows a wave of criticism over Grok’s recent responses, including a string of over-the-top claims portraying Musk as physically superior to elite athletes and even historical figures.
In one viral exchange, Grok selected Musk over NBA star LeBron James in a fitness comparison, arguing that the billionaire’s “sustained grind, managing rocket launches, EV revolutions, and AI frontiers, demands a rarer blend of physical endurance.” The chatbot also described him as the “fittest man alive” and suggested he would defeat former heavyweight champion Mike Tyson in a boxing match.
Musk attempted to dismiss the unusual responses, attributing them to adversarial prompting: “Earlier today, Grok was unfortunately manipulated by adversarial prompting into saying absurdly positive things about me.” He followed with a self-deprecating remark: “For the record, I am a fat retard.”
These incidents have intensified ongoing concerns about the model’s neutrality and safety controls.
Musk announced earlier that Grok 4.1 had received “many updates and fixes” intended to make its outputs more reliable. “Many updates and fixes have been applied to Grok 4.1 and many more to come! Going forward, Grok 4.1 will spend more compute time thinking about your question to improve accuracy.”
He also amplified a video explaining how users can report problematic replies directly within the interface. The instructional post noted that users can tap a thumbs-up to approve an answer, or use the three-dot menu to submit an issue report. Responding to it, Musk wrote: “Critical feedback about Grok is super helpful. We will keep iterating until Grok is the best in every way.”