Chinese military researchers are reportedly developing an AI model for military use based on Meta's AI



It has been revealed that the Academy of Military Sciences, a top Chinese research institute with ties to

the People's Liberation Army , is using Meta's AI model, Llama , to develop AI models that can be used for military applications.

Exclusive: Chinese researchers develop AI model for military use on back of Meta's Llama | Reuters
https://www.reuters.com/technology/artificial-intelligence/chinese-researchers-develop-ai-model-military-use-back-metas-llama-2024-11-01/



Report: Chinese researchers used Llama 13B to build chatbot optimized for military use - SiliconANGLE
https://siliconangle.com/2024/11/01/report-chinese-researchers-used-llama-13b-build-chatbot-optimized-military-use/

Chinese army scientists use Meta technology to create 'military AI'
https://www.telegraph.co.uk/business/2024/11/01/chinese-scientists-use-meta-technology-create-military-ai/

Chinese researchers build military AI using Meta's open-source Llama model — ChatBIT allegedly performs at around 90% of the performance of OpenAI GPT-4 LLM | Tom's Hardware
https://www.tomshardware.com/tech-industry/artificial-intelligence/chinese-researchers-build-military-ai-using-metas-open-source-llama-model-chatbit-allegedly-performs-at-around-90-percent-of-the-performance-of-openai-gpt-4-llm

Report: China Turns Meta Llama Into a Military AI | Extremetech
https://www.extremetech.com/defense/report-china-turns-meta-llama-into-a-military-ai

In a paper published in June 2024 and confirmed by Reuters, a researcher from the Academy of Military Science , an affiliated research institute of the People's Liberation Army of China, introduced an AI model called 'ChatBIT' based on an early version of Meta's AI model, Llama. The research team built a military-specific AI model, ChatBIT, by incorporating their own parameters into Llama-13B, an AI model with 13 billion parameters developed by Meta. The paper states that the purpose of ChatBIT is to 'collect and process information and provide accurate and reliable information for decision-making during military operations.'

The paper explains that 'ChatBIT is optimized for dialogue and question-and-answer tasks in the military field.' 'In the future, with improvements in technology, ChatBIT will not only be applied to information analysis, but also to strategic planning, simulation training, and command decision-making.'

In addition, ChatBIT has about 90% of the performance of ChatGPT, an OpenAI chat AI, and is said to be superior to other AI models. However, the researchers did not provide detailed explanations of how they defined ChatBIT's performance or whether this is actually being used in practice.



'This is the first time that substantial evidence has been presented that military experts from the People's Liberation

Army of China are systematically studying and seeking to leverage open source large-scale language models (LLMs), and Meta's LLMs in particular, for military purposes,' Sunny Chung, an associate fellow at the Jamestown Foundation who specializes in China's emerging and dual-use technologies, including AI, told Reuters.

Meta has released multiple AI models, including LLaMA, but imposes its own restrictions on the use of AI models, such as requiring them to obtain licenses. Meta's terms of use prohibit the use of AI models for 'military, war, nuclear industry, or espionage activities,' as well as the development of weapons or content intended to 'incite or promote violence.'

'The People's Liberation Army's use of Meta's models is unauthorized and violates our Terms of Use,' said Molly Montgomery, Meta's director of public policy.

In addition, a Meta spokesperson said, 'In the global race for AI, given that China has already invested more than $1 trillion (about 150 trillion yen) to overtake the United States in AI, there is no role for a single, outdated version of the American open source model to play,' emphasizing that the AI model China is using is outdated.



Reuters has contacted Chinese officials and authorities for comment but had not received a response at the time of writing.

Reuters explained that it was unable to verify ChatBIT's capabilities or computing power, but noted that 'ChatBIT only incorporates 100,000 military conversation records, which is an overwhelmingly small number compared to other LLMs.' Joel Pinault, Meta's vice president of AI research, also said, 'Compared to most models trained on trillions of tokens, this number (100,000 military conversation records) is very small. I wonder what it is actually achieving in terms of various functions.'

Another academic paper reviewed by Reuters describes how two researchers from

the Aviation Industry Corporation of China , a company with ties to the People's Liberation Army, are using Llama 2 to 'train aerial electronic warfare jamming strategies.' Liberation Army Daily also published a commentary in April 2024 on 'how AI can be used to accelerate research and development of weapons and equipment, aid in the development of combat simulations, and improve the efficiency of military training,' and Reuters reports that there is no doubt that the Chinese government is researching AI for military use.

'Can we keep China out of the cookie jar? No, I don't think we can,' William Hannas, a chief analyst at Georgetown University's Center for Security and Emerging Technologies (CSET), told Reuters, noting that it's difficult to stop China's advances in the field of AI. According to a paper published by CSET in 2023, there are 370 Chinese research institutes that have published papers on general artificial intelligence (AGI). Hannas also said, 'There's so much collaboration between China's best scientists and America's best AI scientists,' making it difficult to exclude Chinese researchers from AI research.



Following this report, Meta announced that it would offer Llama to the US government and contractors working on developing national security applications on November 4, 2024 local time. Specifically, it will partner with companies such as Accenture Federal Services, Amazon Web Services, Anduril, Booz Allen, Databricks, Deloitte, IBM, Leidos, Lockheed Martin, Microsoft, Oracle, Palantir, Scale AI, and Snowflake to provide Llama to government agencies. Meta noted that responsible use of open source AI models 'will help advance security worldwide and position America in the global race for AI leadership.'

Open Source AI Can Help America Lead in AI and Strengthen Global Security | Meta
https://about.fb.com/news/2024/11/open-source-ai-america-global-security/



Related Posts:

in Note, Posted by logu_ii