Google researchers have developed 'Infini-attention', a method that allows Transformer Large Language Models to process text of infinite length.

Infini-attention works by transferring excess data to compressive memory to maintain additional context within the active memory of the model.

The new model has shown a 114-fold increase in memory efficiency over baseline models in tests.

It has outperformed existing LLMs in tasks that require understanding long contexts, such as language modeling and book summarization.

Although currently for research, the technique has potential for widespread use, promising to enhance LLMs' performance and enable novel applications.

The advancement suggests a future for LLMs with significantly improved handling of lengthy texts, optimizing both computational and memory resources.