AI 导读
关于“AI泡沫”的一些餐巾纸数学 显然，未来所有创建的数据都会经过LLM，这比现在多100000倍的数据 所以如果模型成本下降100倍，我们只需要多1000倍的GPU
正文
some napkin math for the "ai bubble"
obviously all data created will pass through an llm in the future, which is 100000x more data than now
so if model cost goes down 100x, we will only need 1000x more gpus
来源：gabriel · x.com