Breakthrough research from the University of California San Diego shows that GPT-4.5 not only imitates but also "beats" humans when 73% of participants identified it as a real person, higher than the rate of accurately identifying real people.
OpenAI's GPT-4.5 and Meta's Llama-3.1 have officially passed the Turing test. Photo: Investopedia
The experiment was carefully designed with 300 volunteers participating in five-minute chat sessions with both AI and real people. Notably, when instructed to act as a young person familiar with online culture, GPT-4.5 demonstrated an astonishing ability to convince. In contrast, the version without the personified information only scored 36%, proving that the "immersion" factor played a key role in this success.
“This is not just about AI passing a test. It raises profound questions about how we define intelligence and cognition,” said lead researcher Dr Cameron Jones, warning that AI could replace humans in short interactions, opening up both opportunities and challenges in terms of job automation and cybersecurity.
This achievement comes at a time when AI is strongly penetrating into life. However, experts such as Google's François Chollet are still skeptical: "The Turing test is a thought experiment, it should not be considered absolute proof of machine intelligence".
As the line between humans and machines continues to blur, a big question arises: Are we ready for a world where machines can mimic humans so perfectly that they are indistinguishable? The answer may well shape the future of humanity in the coming decades.
Cao Phong ( according to HT, Futurism)
Source: https://www.congluan.vn/ai-vuot-qua-bai-kiem-tra-turing-lam-mo-ranh-gioi-giua-con-nguoi-va-may-moc-post341417.html
Comment (0)