| Mobile Web

Repeating the same question twice can lift AI accuracy from 20 percent to 90 percent

Large language models can give different answers to the same prompt depending on how it is written or arranged. Google Research released findings that simply repeating a prompt twice can meaningfully improve accuracy. The method, called prompt repetition, reduces missed information in sequential reading. Tests on seven models found accuracy improved in 47 of 70 conditions and never worsened. Gains were strongest on long or complex inputs, while effects were limited for reasoning-inducing prompts.