Tag
Nathan Lambert and Finbarr Timbers discuss the latest post-training recipes for large language models, including DeepSeek V4, GLM 5.1, Kimi K2.6, and the industry shift to multi-teacher on-policy distillation.