AI can pass the Turing Test in live chats and appear more human than us. I am spooked now

UC San Diego researchers tested live, three-party chats and found GPT-4.5 was judged human 73% of the time when given a persona prompt.

LLaMa-3.1-405B also fooled judges in many cases, showing that top chatbots can already blend into ordinary conversation.

The findings revive Turing Test questions in a real-world setting and sharpen concerns about AI deception and disclosure.

The study, published in PNAS, suggests chatbot trust issues could affect support, education, social interactions, and political messaging.