论文实测多智能体系统与持久记忆的真实收益
一篇给多智能体热潮泼冷水的论文：实测显示持久记忆的准确率提升置信区间跨零，但 KV cache 能省一半多，数字都摆出来了。
这篇论文测试了一个三层 chain-of-agents 系统，把长上下文拆分到多个智能体后，每条查询的峰值 KV cache 降到 14.3 MiB，而单次处理和 RAG 基线约为 35 MiB。持久 trace-recall 层只增加 0.368 MiB 峰值缓存，准确率变化为 +0.015，置信区间跨零，说明收益不显著。作者解释原因：单问题基准里每道题有独立证据且独立评分，存储的 trace 没有可召回的有用内容。论文编号为 arXiv:2610.07782。
Must-read paper on the effectiveness of multi-agent systems and persistent memory. We have recently seen many viral examples of multi-agent systems scaled via persistent memory. This work takes a closer look. They test one three-tier chain-of-agents system. Splitting long-context work across agents cuts peak KV cache to 14.3 MiB per query, against about 35 MiB for single-pass and RAG baselines. The persistent trace-recall tier adds 0.368 MiB of peak cache and changes accuracy by +0.015, with a confidence interval that crosses zero. The authors explain why. Single-question benchmarks give each item its own evidence and score it independently, so stored traces have nothing useful to recall. Paper: arxiv.org/abs/2610.07782 Chat with Paper: academy.dair.ai/papers/persist… 💬 9 🔄 2 ❤️ 10 👀 1379 📊 10 ⚡
