Chinese research institutions with connections to its military reportedly developed AI systems using Meta Platforms’ open-source Llama model to collect and process intelligence.

Reuters reported papers it viewed indicated chatbots developed by researchers using Meta Platforms’ Llama 13B model have military applications, an allegation the social media giant rejects.

The news agency reported six Chinese researchers from three different institutions with connections to the People’s Liberation Army (PLA) released a paper about AI in June.

Reuters stated the researchers used Meta Platforms’ early Llama 13B large language model (LLM) and trained it on military data with the goal of making a tool which could gather and process intelligence, and help make operational decisions.

The ChatBIT model was “optimised for dialogue and question-answering tasks in the military field”, Reuters reported. The research paper indicated the model outperformed other AI models which were around 90 per cent as capable as OpenAI’s ChatGPT-4.

ChatBIT was trained using 100,000 military dialogue records, which a professor told Reuters is a “drop in the ocean” when compared to other LLMs trained with millions of tokens.

Meta responds
A Meta Platforms representative told Mobile World Live “the alleged role of a single and outdated version” of a domestic “open-source model is irrelevant when we know China is already investing more than a trillion dollars to surpass the US technologically”.

“Chinese tech companies are releasing their own open AI models as fast, or faster, than companies in the US.”

Kyle Miller, a research analyst at Georgetown’s Centre for Security and Emerging Technology (CSET), stated on X there is no indication “the model would be very useful for a military, especially compared to newer Chinese models that are better than Llama-13B”.

“The Chinese researchers created a simple Q and A bot that could answer basic questions about the military. Nothing in the paper suggests the model would be very useful for military applications.”

Meta Platforms CEO Mark Zuckerberg is a big proponent of open-source AI.

On a recent earnings call he said it “seems pretty clear to me that open-source will be the most cost-effective, customisable, trustworthy, performant and easiest-to-use option” for developers.