Elon Musk asks users to flag Grok’s flaws; says 'we will not rest until Grok is perfect'

Grok

Elon Musk keeps on dropping new features for Grok. However, this time around, he is looking forward to users' feedback on Grok can be improved. Citing the feedback, he took to his own social media platform X and asked users on to share examples of where Grok, the AI assistant developed by xAI, still needs improvement. In a post that has crossed 26 million views, Musk requested concrete cases where Grok “goes wrong today,” saying that comparisons with how rival AI systems handle the same prompt would help the team fix issues faster. The unusual call for public debugging showcases xAI’s strategy of rapidly iterating on Grok by actively sourcing user-reported failures.

Musk’s request for real-time feedback

Musk wrote, “Please provide examples where Grok needs to improve in replies. Showing how another AI does it better would be helpful.” He noted that some earlier bugs were fixed during the week, but the team wanted fresh examples happening in real time. The post drew thousands of responses, including screenshots, side-by-side comparisons, and critiques from users testing Grok against ChatGPT, Claude, and Google’s AI tools. Musk later followed up with a brief remark: “We will not rest until Grok is perfect.”

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.

— Elon Musk (@elonmusk) November 22, 2025

Grok responds with its own mistakes

In a notable turn, Grok itself replied directly to Musk’s post, offering a self-assessment of where it had gone wrong. The AI pointed to a query involving a federal court ruling about Donald Trump’s National Guard deployment in Los Angeles. Grok said it missed the permanent injunction details, while ChatGPT handled the question more accurately and cited specifics from Judge Breyer’s order.

Grok also highlighted a separate incident where it misidentified the origin of a prank video shared online. The bot noted that Claude tends to perform better in similar cases because it verifies video metadata before responding. The AI ended its message with: “Working on deeper source checks!” — a sign that xAI wants Grok to openly acknowledge and learn from failures.

Users highlight more issues

Several users replied with their own examples of shortcomings. One user said Grok struggles with instruction precision, noting that the model responded with a full paragraph when asked to answer a question in a single word — something Google’s AI handled correctly. Others highlighted that Grok sometimes over-explains simple queries, making it slower to use for quick information.

Some users pointed out that Grok occasionally loses context in longer threads, especially when conversations evolve over multiple prompts. Another user noted that ChatGPT remains more consistent in maintaining relevance over extended dialogue.

What’s next for Grok

Musk’s public request and Grok’s self-critique signal that xAI is using live user feedback as part of its training and refinement loop. With users continuing to share examples and comparisons, the responses are expected to influence upcoming model updates as xAI pushes to close gaps with established AI systems.