Google's AI Gemini Faces Issues With Gibberish Responses; Fixes ...

Google's AI chatbot, Gemini, has been encountering issues, with users reporting instances of strange, nonsensical responses. A Reddit user shared a screenshot showing Gemini replying with jumbled code and characters from various languages, offering no useful information. In the conversation, the chatbot produced a massive chunk of repeating words that didn't address the user's query.

Another example involved a user asking, "If 0 was a Roman numeral, how would it look?" Gemini responded with a lengthy, irrelevant text full of random words and symbols. Additionally, when asked about satellite TV channels, Gemini produced a bizarre string of "Not Cauchy, Cauchy, Cauchy," referencing the mathematical concept of the Cauchy sequence.

These issues, while not widespread, are certainly annoying. However, the AI isn't entirely broken; a simple refresh or rephrasing of the question often resolves the problem. Reports of Gemini's gibberish responses have appeared on both the web and its Android variant.

This isn't the first time Google's generative AI has behaved oddly. Users might recall recent instances where AI Overviews provided erroneous responses, such as suggesting people glue cheese onto their pizza. These outlandish suggestions are believed to stem from the AI's inability to distinguish between factual information and nonsensical or satirical content.

According to Google's Head of Search, Liz Reid, such issues arise from the AI encountering "information gaps" with atypical questions lacking sufficient factual information. The AI tends to pull from user-generated content, which can be humorous or misleading. In response, Google has been working to implement a limiter for its Search AI to reduce the inclusion of satire and humor, aiming to improve the accuracy and relevance of the responses.

Despite these setbacks, Google is actively addressing the issues to enhance the reliability of its AI tools.