A new paper by researchers at Google claims to give large language models (LLMs) the ability to work with text of infinite length. The paper introduces Infini-attention, a technique that configures language models in a way that extends their “context window” while keeping memory and compute …