Some Grok users report gibberish replies on Grok.com

In short: Some people using xAI’s Grok chatbot say it has been answering with gibberish instead of normal text.

What's going on

Several Grok users told TechCrunch that the chatbot has been returning nonsense replies, especially when they ask it to do something specific, like generate a PDF. One user shared a response that looked like random words strung together for multiple paragraphs.

Users who checked what the chatbot was outputting also reported seeing strange strings of links, including links connected to “reinforcement learning” research. Reinforcement learning is one way to train an AI, a bit like teaching a dog tricks using rewards when it does the right thing.

The reports appear to be concentrated among people using “Grok Lite,” and some users said they noticed the problem as early as Wednesday morning. TechCrunch said it could not reproduce the issue in its own testing, which suggests it may affect only a smaller group of users.

The problem also seems limited to direct questions on Grok.com. The Grok account on X.com was not affected.

What to watch

On X, the Grok account acknowledged the issue and called it a “temporary generation glitch,” meaning the system sometimes produces “word salad” by mistake. The account suggested starting a fresh chat or regenerating the answer, and said this often fixes it, although some users reported the gibberish continued even after multiple refreshes. xAI did not respond to TechCrunch’s request for comment.

Source: TechCrunch AI