Updated
Updated · Gizmodo · Aug 20
Grok Spews Gibberish to Users as Satisfaction Trails at 24.6%
Updated
Updated · Gizmodo · Aug 20

Grok Spews Gibberish to Users as Satisfaction Trails at 24.6%

3 articles · Updated · Gizmodo · Aug 20

Summary

  • Users reported this week that xAI’s Grok was answering direct prompts with nonsensical text and, in some cases, linking to reinforcement-learning research sites instead of cited sources.
  • Wednesday complaints centered on Grok Lite, and the official Grok account described the problem as a “rare temporary generation glitch” that can sometimes clear after a refresh.
  • TechCrunch said it could not reproduce the malfunction itself, but multiple users posted examples of “word salad” responses across X and Reddit.
  • The glitch lands as Grok already struggles commercially: a YouGov UK poll put positive satisfaction at 24.6%, second-worst among major AI platforms, while Recon Analytics estimated only 0.174% of users paid for it in Q2 2026.

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?