Grok Lite: Inconsistent Answers, xAI Mentions a Temporary Bug
Le brief IA que les pros lisent chaque soir
Les 7 actus IA du jour, décryptées en 5 min. Gratuit.
Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.
Choisis ton rythme
Gratuit · Pas de spam · Désabonnement en 1 clic
xAI has provided no technical details regarding an outage that caused Grok Lite to generate incoherent text on its web interface. On X, the chatbot mentioned a rare temporary generation bug and suggested refreshing the tab or starting a new conversation, a method that seems to have helped many users. Similar episodes have already affected ChatGPT, Gemini, and Claude.
xAI does not publish any technical details about the outage
xAI has not released any technical information about the origin of the incident, and the exact cause remains publicly unknown. AI companies rarely communicate about such malfunctions, and xAI appears to be following this approach in the case of Grok. Comparable cases have been reported elsewhere: in 2024, ChatGPT experienced an episode of incomprehensible responses to simple queries; Gemini received similar feedback; and Claude went through a period of drift in 2025, particularly among users employing it for coding. All these services rely on models that predict, through mathematical formulas, the most likely continuation of a text; when this prediction fails, the output turns into gibberish. Such failures can occur at any time, with possible causes ranging from faulty code to a mathematical bug.
On the web, Grok Lite generated nonsensical paragraphs
During the week, users of Grok Lite witnessed the bot producing long blocks of nonsense. According to several testimonies published earlier in the week on Reddit, the assistant responded normally before degrading into incoherent and buggy statements. The phenomenon seemed confined to the web interface, without affecting mobile applications. One example of an unreadable response included: “Some are local but added, and security police but correct if search uses glue for already.”
On X, Grok mentions a temporary bug and suggests a workaround
When directly questioned, the Grok account on X described the episode as a “word salad” due to a rare temporary generation bug. It recommended refreshing the browser tab or opening a new discussion to restore normal behavior, a method that seems to have worked for most users. Reports of corrupted responses appear to have decreased since Friday. A new attempt did not reproduce the anomaly.
Brief IA — L'actualité IA en français
L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.