There’s some truth in that, but isn’t making a radically cheaper version also a new idea that deepseek didn’t know whether it would work? I mean, there was already research into distillation, but there was already research into some of (most of?) OpenAI’s ideas.