Every hyperscaler now has an LLM that's less efficient and more expensive and horribly unprofitable versus Deepseek's open source model which can now be adopted and turned into a competitor to GPT4 or o1 except it's cheaper and you know more about how it works. Maybe it's fine? But how?