According to The New York Times, Google’s AI Overview feature has a 91% accuracy rate using Gemini 3, meaning it delivers tens of millions of incorrect answers hourly. Based on Google’s processing of over 5 trillion searches annually, this translates to hundreds of thousands of inaccurate responses per minute.
Oumi’s analysis using the SimpleQA benchmark showed that Gemini 2 achieved 85% accuracy, while Gemini 3 improved to 91%.
Related News
Why do some people think AI will change the world, while others think it’s ordinary? Karpathy’s two diagnoses
Karpathy “Let LLMs argue with themselves”: a 4-step method to counter thinking biases with AI
AI drives the U.S. Q1 GDP growth up 75%, and the top five companies’ capital expenditures in 2027 may exceed $1.1 trillion