Unverified50% confidenceBenchmarkExact time

On update-fact tasks, embedding retrieval methods exhibited unpredictable collapses, with success rates fluctuating between 0.30 and 0.95 and a maximum gap of 0.45 across different random seeds.

1

Sources

50%

Confidence

Long-term

Relevance

9/11/2026

First Seen

Sources

New Standards for LLM Long-Term Memory Evaluation: MERIT Benchmark Reveals the Cost-Performance Tradeoff

Related Entities

嵌入检索concept MERITproduct

Related Claims

Unverified通过记录随机种子可精确复现导致失败的特定顺序，再通过二分法逐步缩小污染源的范围69% similar Unverified撞库攻击利用已知有效凭证批量尝试登录，成功率远高于随机猜测，约0.1%-2%的撞库尝试能够成功69% similar Unverified在某些任务上，仅因随机种子不同导致的性能波动幅度可能超过某些论文声称的方法创新带来的提升68% similar Unverified增加子线程数量不等于增加智力或能力，当数量超过阈值（如20个甚至100个）会出现重复搜索、相互锚定等问题导致效率下降68% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/892958

API

curl https://kongchang.com/api/v1/knowledge/claims/892958

MCP

get_claim(id=892958)