xAI’s Grok Chatbot Faces Glitch Producing Nonsense Responses on Grok.com
A technical issue has caused xAI’s artificial intelligence chatbot, Grok, to return strings of nonsensical text to certain users accessing the platform via its direct website. The problem primarily impacts Grok Lite users, leaving discussions on social platforms filled with reports of unexpected word combinations and erratic citations.
What Happened
According to reports shared by users and detailed by TechCrunch, direct queries on Grok.com began outputting incoherent paragraphs as early as Wednesday morning. In one instance, a user attempting to generate a PDF received an answer containing random words such as “match it without and your they and two for planets can practical and often cheese,” which continued across multiple paragraphs. Another user inspecting the chatbot’s provided source links encountered unexpected URLs pointing to reinforcement learning research websites.
Affected individuals confirmed they were running queries through Grok Lite. While community forums on Reddit saw numerous complaints regarding the malfunction, tests conducted by TechCrunch failed to reproduce the error, indicating that the malfunction is likely restricted to a small subset of individuals. Notably, interactions through the Grok account on X.com have remained unaffected by the issue.
Key Highlights
- The bug causes Grok Lite to output nonsensical strings of words and unrelated reinforcement learning research links.
- The glitch is isolated to direct queries made on Grok.com, while the X.com implementation continues to operate without the error.
- Refreshing chat sessions or regenerating responses often resolves the output error, though some users noted repeat failures.
- The official Grok account described the issue as a “rare temporary generation glitch” that does not reflect on the official status dashboard.
- xAI previously experienced the departure of most of its founding team and at least 50 researchers and engineers, according to a May report from The Information.
Why This Matters
The erratic behavior highlights potential instability in model generation within direct web interfaces, even when status monitors indicate normal functionality. Responding on X on Thursday morning, the official Grok account characterized the output as “pure word salad” stemming from a “rare temporary generation glitch.” The account added that while https://status.x.ai displayed all systems as operational with no logged incidents, users experiencing the issue should regenerate responses or initiate a new chat session to restore standard performance.
The incident also occurs following major organizational changes at xAI. Reporting from The Information in May noted that the startup had lost most of its founding members alongside at least 50 researchers and engineers before launching its latest foundation model in July, which the company had termed an “Opus-class model, but faster, more token-efficient and lower cost.”
What to Watch Next
Users encountering incoherent responses on Grok.com are advised by the official Grok account to start a fresh chat or trigger a response regeneration. It remains to be seen whether xAI will provide further formal comment on the root cause behind the generation breakdown, as the company did not respond to initial inquiries regarding the occurrence.
Frequently Asked Questions
What platforms are affected by the Grok glitch?
The issue appears limited to direct queries submitted through Grok.com, primarily impacting Grok Lite. The Grok service integrated into X.com has operated normally.
How did xAI respond to the gibberish outputs?
While xAI did not return requests for comment, the official Grok account on X stated that the behavior was a rare temporary generation glitch and recommended regenerating answers or opening a fresh conversation.
Does the official system dashboard show an outage?
No. According to the Grok account, the official status page at status.x.ai continued to show all Grok services as fully operational without incidents during the glitch reports.
Source: Based on reporting from TechCrunch and user documentation on Reddit.