Two common chat shoes have reached an important line-they have passed the Turing test, scientists from the University of California in San Diego (USA) said. We are talking about GPT and LLAMA models. This may mean robots have become as smart as everyone.
Testing, created in 1950, was used to check the intelligence of machines. If the researchers could not distinguish the result of a person's result from a person's results, it was believed that the machine had surpassed the test.
The team tested four artificial intelligence models: released in February 2025 GPT-4.5, the previous version called GPT-4O, LLAMA Model and the 1960 Eliza chat program. The first three are the big language models of the Viking people – deep algorithms can recognize and create text based on large data sets.
Experts have attracted 126 students from the University of California and 158 people from the full data collection platform. Participants have led people online with another person or artificial intelligence, and they don't know who is their dialogue. Therefore, GPT-4.5 has been taken to one person in 73% of cases more often than real people. LLAMA-3.1 is recognized as one of 56% of cases. In the basic models, Eliza and GPT-4O, indicators are 23% and 21% respectively.
AI models show the best results if they are instructed in advance, how to impersonate a person. But this does not mean that they cannot pass the test without suggestions. The authors emphasize that this is the first case of success by the artificial intelligence of the Turing test, the Arxiv report.
Earlier The nerve network was first recorded in the university when he was a student. The leadership of the Applied Art University in Vienna has taken such a step. A model called Flynn has been recorded in the digital art program. She passed the standard application process. Leaders note that there is no written standard, whereby students must be one person.