Grok keeps sending gibberish responses to users

An unexpected technical malfunction has left many users of xAI’s Grok chatbot baffled, as the AI has begun outputting incoherent strings of text instead of standard responses. Reports of the glitch began surfacing as early as Wednesday morning, primarily impacting those utilizing the Grok Lite service.

A Breakdown in Communication

The nature of the errors varies, but the results are consistently nonsensical. One user attempting to generate a PDF was met with a bizarre string of words, including phrases like “match it without and your they and two for planets can practical and often cheese.” Other users reported that the chatbot provided lists of links pointing toward reinforcement learning research papers rather than answering their direct queries.

While the issue has caused significant frustration across the Grok community on Reddit, it does not appear to be universal. Testing conducted by industry observers failed to replicate the error, suggesting that the bug is currently limited to a specific subset of the user base.

Official Response and Troubleshooting

The official Grok account on X (formerly Twitter) addressed the situation on Thursday morning, characterizing the incident as a transient technical failure.

“That pure word salad is a rare temporary generation glitch. Official status at https://status.x.ai shows all Grok services fully operational with no incidents. Start a fresh chat or regenerate—it usually clears right away. Sorry about the gibberish.”

For those experiencing the issue, the company suggests the following steps:

• Refresh the session: In many cases, starting a new chat window resolves the output errors.
• Regenerate the response: If the gibberish persists, triggering a new generation attempt may clear the cache.
• Monitor the status page: Users can verify if the service is experiencing broader outages at the official xAI status portal.

Contextual Challenges at xAI

The glitch arrives during a period of transition for the Elon Musk-backed AI firm. Recent reports from The Information indicate that xAI has faced notable internal instability, including the departure of a significant portion of its founding team and the loss of at least 50 researchers and engineers.

Despite these internal shifts, the company continues to iterate on its technology. In July, xAI unveiled its latest foundation model, touting it as an “Opus-class” system designed to be more efficient, faster, and more cost-effective than its predecessors. For now, users are left waiting to see if these "word salad" errors are fully resolved or if they signal deeper stability concerns within the platform.