Chinese AI lab DeepSeek debuts DeepSeek-R1, an MIT-licensed model that does well with math, code, and reasoning tasks, alongside other open and distilled models (Simon Willison/Simon Willison's Weblog)