August M
Developing cutting-edge artificial intelligence usually demands massive supercomputers alongside years of costly research.
When high-tech hardware is difficult to obtain, clever workarounds can quickly become invaluable. Nations seeking a military advantage are now finding new ways to train defense systems using borrowed computing power, reports Reuters.
Clever military shortcuts
Chinese military researchers are tapping top American artificial intelligence platforms to upgrade domestic defense systems. According to a report by Reuters based on academic papers, scientists affiliated with the People’s Liberation Army are extracting outputs from models created by OpenAI and Anthropic.
The strategy relies on a popular tech method known as model distillation. Instead of building massive software from scratch, developers use answers generated by powerful foreign tools to instruct smaller, highly tailored programs.
This method enables Chinese teams to sidestep Washington’s strict export bans on advanced computer chips. By running condensed AI on local hardware, researchers can deploy smart tools on military devices without relying on massive data centers.
Sunny Cheung, a research fellow at the Jamestown Foundation, noted that scientists want more than just basic facts. “Teaching a model the right answer is one thing but teaching it the reasoning behind the answer is much harder,” Cheung said. “These papers show Chinese military-linked researchers are trying to transfer that expensive, proprietary reasoning from Western models into smaller systems they can control and deploy locally.”
Drones and code
The reviewed documents reveal concrete battlefield uses. Analysts with a military intelligence unit in Beijing used OpenAI’s GPT-3.5 to process sensitive software code, creating condensed summaries to construct a domestic model for secure internal networks.
At defense institutions, researchers applied distillation to aerial drones and naval equipment. One project compressed visual data tools so unmanned aircraft could analyze live video and navigate during total communication blackouts.
Despite these rapid gains, tech specialists emphasize that stripped-down models cannot completely replace frontier software.
Trevor Koverko, co-founder of AI data firm Sapien, explained that distilled systems remain limited. “It is best understood as transferring selected capabilities into a cheaper, locally controlled system, not achieving independence from frontier AI,” Koverko stated.
Sources: Reuters, Jamestown Foundation
