CSIS study highlights DeepSeek, another Chinese model, as more ‘hawkish’ AI

The Center for Strategic and International Studies has run China’s DeepSeek AI model through a benchmark it created to assess large language models for biases in coming to foreign policy decisions, and found the results to be part of a concerning pattern of certain models recommending escalation in reaction to international disputes.

“Overall, our evaluation reveals DeepSeek shares a troubling tendency toward more hawkish, escalatory recommendations seen in other Chinese LLMs like Qwen2,” CSIS wrote in an April 16...