AI 导读
我们还将这些趋势外推到了高达 1000 万 token 的上下文。在这样的长度下，GPT 曲线的二次方缩放会变得明显得多，使这些模型的使用可行性大幅降低。https://t.co/0Krypqe9Zn
正文
We also extrapolated these trends to contexts of up to 10 million tokens. At these lengths, the quadratic scaling of the GPT curves would become much more noticeable, making these models far less feasible to use. https://t.co/0Krypqe9Zn
来源：@EpochAIResearch · x.com