Reuters probe finds Chinese labs distilling US AI models to train homegrown rivals
Reuters (via MarketScreener) · · AI-summarised by FinFlash
A Reuters review of more than 300 Chinese academic papers, patents, corporate filings, hiring notices and government documents found Chinese researchers using outputs from US AI models to train local rivals. One 2024 Shanghai Jiao Tong University paper took about 5,000 answers from OpenAI's o1 to build 10,750 training examples for Alibaba's Qwen2.5-72B. A US official called industrial-scale distillation the main reason Chinese models have narrowed the gap, though two sources credited engineering instead.
So What?
Adds weight to Washington's push for tighter AI controls on China ahead of the Trump-Xi meeting; watch Chinese AI names such as Alibaba for regulatory headline risk.
Understand the context
Published information may change. Summaries can contain errors; check the original source. Educational information only, not investment advice.