Tag
Researchers introduce BEHEMOTH benchmark and CluE cluster-based prompt optimization to enable LLMs to extract and retain heterogeneous memory across diverse tasks, achieving 9% gains over prior self-evolving frameworks.