Researchers from Huawei and University College London have developed a new approach called EM-LLM, which integrates aspects of human episodic memory and event cognition into large language models (LLMs). This allows LLMs to have infinite context lengths while maintaining their regular functioning.