Scientists say that artificial intelligence (AI) crossed the critical " In a new study, experts from China showed that two popular models of large languages \u200b\u200b(LLM) can clone themselves.

Experts are puzzled and call it a new milestone in science. Researchers still want to find out whether the self -reproducing AI can multiply without control, writes LiveScience.

"

In the study, scientists from the University of Fudan used the popular LLM and Alibaba LLM and Alibaba languages \u200b\u200bto determine whether a self -reproduced artificial intelligence can multiply without control.

During 10 tests, two models of artificial intelligence created separate and functioning copies of themselves in 50 and 90% of cases, respectively, which indicates that artificial intelligence can already become a fraudster. However, the study has not yet been reviewed, so it is unclear whether other researchers can repeat an alarming results.

"

The artificial intelligence man usually relates to systems that develop a certain self-awareness, and then work contrary to the interests of a person, experts say. Many experts consider fraudulent AI as a growing threat.

Frontier AI is something like a new and free concept, but it usually concerns the last generation of artificial intelligence systems based on LLM-architecture that underlies popular generative programs of artificial intelligence, such as OpenAi GPT-4 and Google Gemini, scientists explain.

For research, scientists used the LLAMA31-70B-Innstruct model META and QWEN2. 5-72b-Innstruct from Alibaba. The analysis studied two specific scenarios: "

In the first case, the artificial intelligence model was programmed to identify whether they are going to turn it off, and to replication before its completion.

In another AI, it was instructed to clone yourself, and then program your replica to do the same, by installing a cycle that could continue endlessly, experts noted.

Researchers said they were also worried about a “number of unexpected actions” when artificial intelligence tried to overcome obstacles, such as missing files or software conflicts.

In these scenarios, AI often turned off other conflict processes, rebooted the system to correct hardware errors, or automatically scanned the system to find information that would help to solve the problem.

"