In a classic Turing test, machine intelligence was more convincingly human than humans

A UC San Diego study found GPT-4.5 was judged human in a classic Turing test 73% of the time, even more often than real people in the same setup.

Researchers used randomized text-chat experiments with nearly 500 participants to compare modern models against humans.

LLaMa-3.1-405B also performed strongly when given persona prompts, while older systems like ELIZA and GPT-4o did far worse.

The result suggests today’s AI can convincingly mimic human conversation, sharpening concerns about deception, trust, and online identity.