Singapore’s primary Chinese language newspaper Lianhe Zaobao recently reported that, in a paper published in June this year, six Chinese military researchers detailed how they used an early version of Meta’s Llama large language model (LLM) to build their ChatBIT model. These researchers named on the publication are from the Military Scientific Information Research Center of the Chinese Academy of Military Sciences, the National Defense Science and Technology Innovation Institute, the Beijing Institute of Technology, and the Minzu University of China.
The Chinese researchers utilized the Llama 2 13B large-scale language model released by Meta in February 2023. Combined with their own parameters, they built a military-focused artificial intelligence tool for collecting and processing intelligence to provide accurate and reliable information for combat decision-making. The paper states that, after fine-tuning, ChatBIT was optimized for question-and-answer tasks in the military field and that its performance exceeded other artificial intelligence models.
Reportedly, the Chinese models are approximately 90 percent as capable as OpenAI’s ChatGPT-4. The Chinese researchers did not elaborate on how they measured performance or on whether the AI model is already in use in the field.
Meta has publicly released many AI models, including Llama, and has imposed licensing restrictions on the use of these models. Specifically, Meta prohibits the use of its models for “military, war, nuclear industry and espionage.” Meta’s public policy director Molly Montgomery said “Any use of our model by the Chinese People’s Liberation Army is unauthorized and violates our acceptable use policy.”
A fierce debate is ongoing in the U.S. national security and technology community over the consequences of technology companies such as Meta making their models public.
Source: Lianhe Zaobao, November 1, 2024
https://www.zaobao.com.sg/news/china/story20241101-5283840