Notes on DeepSeek

DeepSeek operates out of an unmarked Hangzhou building with only 300 employees, roughly one-tenth the size of Anthropic, and has no current plans to scale headcount. The team does no red teaming, views R1 as their peak achievement to date, and is comfortable staying approximately six months behind U.S. frontier labs. China's AI regulatory model targets downstream applications rather than models directly, and the DeepSeek team's primary societal concern is youth unemployment, not AGI risk.