Meta's Llama AI model turned into a military tool in China

Chinese research institutions linked to the People’s Liberation Army (PLA) have reportedly adapted Meta’s open-source Llama AI model to create a tool for military applications, as seen in academic studies and expert analysis.

In June, researchers from PLA-affiliated institutions, including the Academy of Military Science, outlined how they used an early Llama version as the foundation for "ChatBIT."

By incorporating specific parameters, Reuters reports, they built ChatBIT to gather intelligence, aid in decision-making, and conduct military-oriented dialogue. According to the paper, ChatBIT performed well in tasks like question-answering within a military context, though researchers didn’t specify its real-world use.

Sunny Cheung, an expert from the Jamestown Foundation, observed that this research marks a shift, showing PLA scientists systematically adapting open-source models like Llama for military purposes.

Meta has shared many of its AI models publicly to encourage open innovation, with some restrictions, including bans on military use. However, as open-source models, enforcing these rules proves challenging.

In response, Meta affirmed that PLA usage is unauthorized and incompatible with its policies. The company argued that while its models promote transparency, China’s substantial investment in AI surpasses the potential value of any single open-source model.

Chinese researchers involved include scientists from the Academy of Military Science, Beijing Institute of Technology, and Minzu University.

They suggest that ChatBIT could eventually assist in strategic planning, simulations, and command decisions as its technology advances. Currently, however, ChatBIT relies on a dataset of 100,000 military dialogues – significantly smaller than those used in other large-scale models. Joelle Pineau, Meta’s VP of AI Research, questioned ChatBIT’s capabilities due to its comparatively small dataset, noting the importance of vast data to improve model reliability.

China's advancements in AI, including domestic military and security applications, have raised concerns in the US, where President Biden recently signed an executive order to regulate AI developments. This week, Washington proposed rules to control investment in Chinese AI and other sectors deemed critical to US national security.

According to the Pentagon, open-source models offer both opportunities and risks, and it is closely monitoring advancements by global competitors. William Hannas of Georgetown University's Center for Security and Emerging Technology remarked that as China accelerates AI research, restricting access will be increasingly challenging due to ongoing collaboration between Chinese and Western scientists.

Yup, it's an entropy thing – the toothpaste has left the tube!