China Develops Meta Llama Model-Based Military AI, Pentagon Strictly Monitors

JAKARTA China's top research institute related to the People's Liberation Army (PLA) has used Meta's Llama model, which is publicly available, as a basis for the development of AI tools for military applications. Based on three academic papers and analyses reviewed by Reuters, PLA researchers used Meta's 13B AI Llama model and integrated specific parameters to develop a tool called "ChatBIT."

The tool is designed to support decision-making in a military context, improve intelligence capabilities, and provide accurate and reliable information for military operations.

In a paper published in June 2024, six researchers from three institutions, including two institutions under the PLA such as the Academy of Military Sciences (AMS), explained that ChatBIT was developed from the optimized 13B Llama model for special tasks such as dialogue and questions in the military field.

Based on the test results, ChatBIT was found to be able to exceed the performance of several other AI models, reaching a level of capability of about 90% of ChatGPT-4. However, in its publication, the researchers did not explain in detail the performance metric used or confirm whether the AI model had been implemented in operational services.

Researchers from AMS involved in the development of ChatBIT include the Guotong and Li Weiwei Gangs from the AMS Military Sciences Research Center and the National Institute for Defense Technology Innovation, in collaboration with researchers from the Beijing Institute of Technology and Minzu University.

In their paper, the researchers mentioned that ChatBIT has the potential to be further developed so that it can be applied to intelligence analysis, strategic planning, simulation training, to support future command decisions.

Meta spokesperson Molly Montgomery stated that the PLA does not have a permit to use the Llama model for violating the use policies implemented by Meta. Meta imposes restrictions on the use of the Llama model, including bans on military applications, the nuclear industry, intelligence activities, and arms development. However, as this model is open, Meta faces challenges in implementing the policy effectively.

"The use of our model by the PLA is invalid and contrary to our use policies," Montgomery said, quoted by VOI from Reuters. In addition, Meta emphasizes the importance of open innovation in the midst of global AI competition. "In global AI competition, the role of one old version of the open-source model is irrelevant, especially as China has invested trillions of dollars to surpass the US in the AI sector," Meta's representative added in a written statement.

Meanwhile, the Pentagon stated that it is monitoring the development of a competitor AI capabilities very closely. Pentagon spokesman John Supple stressed that although the open-source model has advantages, there are also weaknesses in terms of security that need to be considered.

China's move to adapt Western AI models such as Llama has also expanded to the domestic security sector. Llama has been used for applications such as "intelligence polling" or intelligence surveillance to process large data and support police decisions.

Some observers say that China's steps in local AI development are increasingly narrowing the gap between technology and the US. According to William Hannas, the lead analyst at the Center for Developing Technology and Security (CSET) at the University ofturbation, there are more than 370 institutions in China whose researchers have published papers related to general artificial intelligence. This supports China's national strategy to become global AI leader by 2030.

"Can we really deter them (China) from accessing this technology? No, I think it's very difficult," Hannas said. He added that collaboration between China scientists and leading US AI scientists makes it increasingly difficult to limit access to technology.