and peers notwithstanding—s1 is instead “fine-tuned" on the pre-existing Qwen2.5 LLM, produced by Alibaba, China’s other top-tier AI lab. Before s1’s training began, in other words ...