Future published study or official press release from OpenAI

UC San Diego Study: GPT-4 Judged Human 54% of the Time in Turing Test

May 16, 2024, 08:58 AM

A new study conducted by researchers at UC San Diego, led by C R. Jones and B K. Bergen, has found that GPT-4, an advanced AI language model, was judged to be human 54% of the time in a pre-registered Turing test. This study is being described as the first robust empirical demonstration that any artificial system passes an interactive 2-player Turing test. However, humans were identified as human 67% of the time, suggesting that GPT-4's performance, while significant, does not fully pass the Turing test. The study involved humans conversing with GPT-4, GPT-3.5, ELIZA, or another human.