I see people claiming that the DeepSeek team thought outside the box. No, what they did was to ensure they were thinking inside the right box, which was a different one than OpenAI and its ilk were using.