Institutions linked to the People’s Liberation Army of China (PLA) have reportedly utilized a publicly available neural network from Meta to develop a tool potentially suitable for military purposes, according to Reuters, citing sources and scientific papers.
Experts based their work on an early version of Llama 13B, applying their own parameters to create what they call ChatBIT—a tool capable of gathering and analyzing intelligence data for military purposes, as well as providing accurate and reliable information for operational decision-making.
ChatBIT is “calibrated and optimized” for military dialogue and “outperforms some other AI models,” the agency claims. However, the method for determining the performance of the neural networks was not specified.
“For the first time, substantial evidence has been obtained that PLA military experts in China are systematically exploring and attempting to exploit the capabilities of open LLMs, especially from Meta, for military purposes,” commented Sunny Chung, a junior research fellow at the Jamestown Foundation.
In July, Meta unveiled a collection of new AI systems, Llama 3.1, including the “much-anticipated” 405B—the “first open-source model of an advanced level.” Although the company prohibits the use of its products for military purposes, the public availability of neural networks limits the ability to impose restrictions.
“In the future, thanks to technological advancements, ChatBIT will be applied not only for information analysis and intelligence. [China] will explore strategic planning, simulations, and command decision-making,” according to a document reviewed by journalists.
In June, it was reported that China had developed an AI commander for military simulations.
Previously, a group with “clear ties” to China, SweetSpecter, attempted a phishing attack on OpenAI employees.