Google AI Overviews generate millions of lies

A recent analysis suggests Google's AI Overviews, powered by Gemini, are producing millions of inaccurate answers daily. While Google claims 90% accuracy, this implies a significant number of falsehoods are presented to users. The study, conducted with AI startup Oumi using the SimpleQA benchmark, found accuracy improved to 91% after a Gemini update, but this still translates to widespread misinformation. Google disputes the methodology, citing potential inaccuracies in the benchmark itself and arguing it doesn't reflect real-world search queries. The company also notes that different AI models are used for various queries to optimize speed and cost, potentially impacting consistency.