Major US companies accused Chinese tech firms of using AI-powered underhanded tactics to boost their own AI models.
“Distillation” techniques allow weaker AIs to train on more sophisticated ones’ outputs; OpenAI told the US Congress that DeepSeek — which shocked the world with its cheap and effective R1 model last year — used unauthorized distillation to boost its models’ capabilities. That could allow DeepSeek to get the benefit of OpenAI’s expensive training without the investment, and potentially undermine US firms’ advantage over Chinese ones.
US companies are also bolstering their defenses against what they consider to be Chinese piracy: Disney sent a cease-and-desist order to ByteDance, accusing it of training its AI model on copyrighted characters.