#依赖推理

MEME Benchmark Reveals Critical Flaws in LLM Memory Systems: Dependency Reasoning Accuracy Below 50%

MEME benchmark first evaluates LLM memory dependency reasoning across 6 major systems—best accuracy only 42%. Deep analysis of cascade and missing task failures reveals path to next-gen AI memory.