GLM-5.2 13-minute critique-to-deployment cycle sets a benchmark for AI knowledge production speed that traditional academic publishing cannot approach