我昨天和一家初创公司交流时,他们说为了让信息能放进大语言模型的上下文窗口,不得不把某些内容进行压缩。但“压缩”本身就是一种“理解”,而在被压缩后的形式中,这些信息反而能被用于其他新的用途。 I talked to a startup yesterday that had been forced to compress certain information to fit it into an LLM context window. But compression is understanding, and in the compressed form the information could be used for other, new things.

Paul GrahamCo-founder of Y Combinator

观点原文

我昨天和一家初创公司交流时,他们说为了让信息能放进大语言模型的上下文窗口,不得不把某些内容进行压缩。但“压缩”本身就是一种“理解”,而在被压缩后的形式中,这些信息反而能被用于其他新的用途。

I talked to a startup yesterday that had been forced to compress certain information to fit it into an LLM context window. But compression is understanding, and in the compressed form the information could be used for other, new things.

AI

相关观点