This post is also available in:
עברית (Hebrew)
In a surprising turn of events, China is reportedly using Meta’s publicly accessible Llama 2 LLM to develop an indigenous AI tool aimed at defense and law enforcement applications. This development has raised concerns among analysts about the potential implications for military advancements in the region.
Recent academic papers reveal that prominent Chinese research institutions linked to the People’s Liberation Army (PLA) are leveraging Meta’s Llama model to create a specialized AI tool named “ChatBIT.” A June paper reviewed by Reuters details how six researchers from three institutions, including two associated with the PLA’s Academy of Military Science (AMS), utilized an earlier version of Llama as the basis for their project.
The researchers fine-tuned the model to focus on military applications, aiming to enhance intelligence gathering and processing capabilities for operational decision-making. According to the findings, ChatBIT demonstrated superior performance in dialogue and question-answering tasks, reportedly achieving around 90% of the capabilities of OpenAI’s ChatGPT-4.
Sunny Cheung, an associate fellow at the Jamestown Foundation, noted that this marks the first substantial evidence of PLA experts systematically exploring open-source LLMs for military use. The research team included individuals from the AMS’s Military Science Information Research Center, the National Innovation Institute of Defense Technology, the Beijing Institute of Technology and Minzu University.
Looking ahead, the researchers indicated that ChatBIT could evolve beyond intelligence analysis to encompass strategic planning, simulation training, and command decision-making. However, they acknowledged that their model was trained on only 100,000 military dialogue records, a rather small dataset compared to other LLMs.
The implications of this development extend beyond military applications. Analysts express concerns that China’s advancements in indigenous AI could enable it to bridge the technological gap with the United States. For instance, two researchers from the Aviation Industry Corporation of China (AVIC) detailed in another paper how they used Llama 2 to develop strategies for airborne electronic warfare, according to Reuters.
China’s integration of Western-developed AI technologies is also evident in domestic security. A June publication discussed using Llama for “intelligence policing,” analyzing large data sets to enhance police decision-making. Additionally, the PLA Daily has highlighted AI’s potential to accelerate weapon development and improve military training efficiency.
Meta, for its part, has made its AI models publicly available but imposes certain restrictions. For example, the company prohibits the use of its models for military purposes, espionage, and activities that could promote violence. However, due to the public nature of these models, enforcement of these provisions remains difficult.
