xAI’s Grok Chatbot Suffers Bizarre Gibberish Glitch

Many users have reported that xAI’s Grok chatbot is producing nonsensical responses due to an unusual malfunction.

For instance, one user, attempting to prompt the model to create a PDF, was met with a response like “match it without and your they and two for planets can practical and often cheese…” followed by multiple paragraphs of similar incoherent text.

Similarly, another user examining source links discovered a series of URLs directing to reinforcement learning research websites.

Users experiencing this problem informed TechCrunch that they were utilizing Grok Lite and first observed the anomalies on Wednesday morning. TechCrunch’s own tests failed to replicate the problem, suggesting it might be limited to a small group of users. xAI has not yet provided a statement regarding the situation.

Despite this, the incident has left numerous Grok users bewildered. The chatbot’s Reddit forum has seen an influx of complaints. While initiating a new chat often resolves the issue, some users reported that the nonsensical outputs persisted even after several attempts to refresh.

This flaw seems to be isolated to direct interactions on Grok.com, as the Grok presence on X.com remains unaffected.

In a reply to disgruntled users, the official Grok account acknowledged the problems on X (formerly Twitter):

var playerInstance_jwplayer_6a87469596feb = jwplayer( “jwplayer_6a87469596feb” );playerInstance_jwplayer_6a87469596feb.setup({playlist: “https://cdn.jwplayer.com/v2/media/xqUApuDZ”,});

“The complete word salad is a rare, transient generation error. Our official status page at https://status.x.ai indicates that all Grok services are fully functional with no active incidents. Please try starting a new chat or regenerating the response—this typically resolves the issue immediately. We apologize for the nonsensical output,” the account posted on Thursday morning.

In related news, xAI has experienced considerable personnel changes recently, with reports indicating the departure of the majority of its core founding members and at least 50 researchers and engineers, as detailed by The Information in May. The company’s latest foundational model, released in July, was touted as “an Opus-class model, yet more rapid, token-efficient, and economical.”

Leave a Reply