Updated
Updated · TechCrunch · Aug 20
Grok Chatbot Spews Gibberish to Users, Hitting a Small Subset on Grok.com
Updated
Updated · TechCrunch · Aug 20

Grok Chatbot Spews Gibberish to Users, Hitting a Small Subset on Grok.com

1 articles · Updated · TechCrunch · Aug 20

Summary

  • Many Grok Lite users reported Thursday that direct queries on Grok.com were returning multi-paragraph nonsense instead of answers, including one PDF request that produced "word salad."
  • User checks of source links pointed to reinforcement-learning research sites, and refreshing the session often fixed the problem, though some said repeated refreshes did not restore normal responses.
  • Grok’s account on X acknowledged a "rare temporary generation glitch" and advised users to start a fresh chat or regenerate, even as xAI’s status page showed all services operational.
  • The issue appeared limited to Grok.com chats rather than the Grok account on X, adding to user confusion as Reddit complaints piled up and xAI did not respond to TechCrunch.

Insights

Why did Grok suddenly start speaking in tongues, and does it reveal a deeper flaw in xAI's rushed development?
Could a simple PDF request exposing internal research links hint at severe security vulnerabilities lurking within Grok's architecture?