Image: UC San Diego TodayUC San Diego study says advanced AI can pass a classic Turing test
• University of California San Diego researchers said a modern AI system passed a rigorous three-party Turing test, with GPT-4.5 judged human 73% of the time in live chats. • The study found LLaMa-3.1-405B was picked as human 56% of the time, while baseline systems ELIZA and GPT-4o were chosen only about 23% and 21% of the time. • Lead author Cameron Jones said the right “persona” prompt let models show “tone, directness, humor and fallibility,” raising questions about deception, trust and how people judge AI online.
today.ucsd.edu