[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-naive-rag-beats-complex-agentic-memory":3,"topics-all":38,"news-related-aa294551-fbef-4cf4-b466-1928ab1925bb":57},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":24,"news_slug":31,"published_at":32,"created_at":33,"modified_at":34,"is_published":35,"publish_type":36,"image_url":14,"view_count":37},"aa294551-fbef-4cf4-b466-1928ab1925bb","朴素 RAG 拿下 6\u002F7 组评测:复杂记忆系统被比下去了","一篇 arXiv 论文把智能体记忆系统拉上考场对照:只做稠密检索加重排、直接读原始对话历史的朴素 RAG,在 LoCoMo、LongMemEval、MEME、MemoryAgentBench 四个基准的七组设定里赢下六组,最多领先 5.53 分。论文同时给出反证,给朴素基线加工记忆反而更慢更差。","过去两年，智能体记忆几乎成了 AI 工程圈最热的方向。Mem0、MemMachine、Memora、EverMemOS——架构越画越复杂，功能越堆越多。\n\n但一篇 10 月 22 日挂上 arXiv 的论文把这堆东西拉进考场跑了一圈，结论有点扎人：最朴素的 RAG 方案，几乎全赢了。\n\n论文是《Is Complex Agentic Memory Always Needed? Naive RAG Is Stronger Than We Thought》（[arXiv:2610.30755](https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.30755)），西安交通大学六位作者完成。对照组 Raw-RAG 极其朴素：不做抽取、不做结构化、不用 LLM 改写历史，只把原始对话切窗口、稠密检索加重排，把证据塞给一个 reader 模型。\n\n## 赢了多少\n\n四个基准——LoCoMo（1540 道题）、LongMemEval（500 例）、MEME（1188 道阶段题）、MemoryAgentBench（2071 题）——对照 12 个记忆系统（LoCoMo 与 LongMemEval 一侧 12 个，MEME 与 MemoryAgentBench 一侧 7 个）。七组「基准×模型」设定里，Raw-RAG 赢下 6 组，只在 MEME 配 DeepSeek-V4-Flash 那组输给 MemMachine。差距见[官方对照表](https:\u002F\u002Fgithub.com\u002FLitmeb\u002FRAW-RAG)：LoCoMo 上 GPT-4o-mini 读手 89.07 对 MemMachine 的 88.12，MemoryAgentBench 上 53.41 对 Memora 的 47.88。摘要的说法是在 6\u002F7 组设定里领先最强对照系统 0.13 到 5.53 分。LongMemEval 那组只赢 0.13 分，基本踩线。\n\n## 更狠的是反证\n\n只是「简单方法也能打平」，顶多算持平论文。真正有杀伤力的是后面三组实验。\n\n第一组审计 Mem0 的写入决策。LoCoMo 的 33902 次写决策里，UPDATE 和 DELETE 只占 4.39%，LongMemEval 的 428858 次里占 2.46%，真正落地的有效变更只有 1.90% 和 1.34%。把 UPDATE 换成纯追加的 ADD、DELETE 换成 NONE，LoCoMo 准确率从 67.36% 升到 69.42%——显式改写记忆没带来任何收益。\n\n第二组给 Raw-RAG 加工记忆，三种方式全部变差：抽原子事实掉到 28.70 分（比基线低 19.62），加摘要掉 5.81 分，图结构扩展掉 5.78 分。论文的解释是抽原子事实会把测试时学习从 47.68% 打到 15.37%，可用示例的格式结构被拆散了。\n\n第三组是成本。同一段 10 集 110 问的 MEME 切片上，Raw-RAG 写入加检索合计 305.91 秒，Mem0 要 4142.95 秒，EverMemOS 要 9301.48 秒。论文的说法是比下一个快 5.1 倍、比 EverMemOS 快 30.4 倍。而加上那三种加工，代价是上亿个 LLM token 和数小时耗时，Raw-RAG 同样任务只要 12 分钟。\n\n## 为什么\n\n机制落在两个指标上：覆盖率（证据够不够回答）和利用率（证据够了之后答对没答对）。MemoryAgentBench 上覆盖率跨度 19.07% 到 45.29%，利用率跨度只有 80.00% 到 89.71%。真正拉开差距的不是「拿到证据后多会读」，而是「该给的证据有没有给到」。论文还引入「路径覆盖」解释长链条问题：MEME 追踪和级联题上，Raw-RAG 条件路径覆盖 62.34%，MemU 只有 9.62%。\n\n最锋利的一刀是完整存储对比。给加工过的记忆库开全部权限，Raw-RAG 和 MemMachine 恢复到 57.54% 和 57.46%，原始证据本身是 61.60%。去掉打包上限后，七个系统全部仍比原始证据低 7.58 到 29.38 分，配对 95% 置信区间都不含零。结论很直白：写入阶段丢掉的信息，靠检索补不回来。\n\n## 谁该紧张\n\n这不是「记忆系统没用」。论文划了三条边界：证据只在离线对话和公开基准上成立；原始存储随历史线性增长，牵涉隐私；状态重建效果还取决于 reader 模型能力。附录里跑 PersonaMem-v2，Raw-RAG 拿 49.62%，七个系统里只排第三。但验收标准是实打实的：改进型记忆系统应当证明自己赢过一个配置良好的朴素 RAG 基线，并说清楚多出来的复杂度贡献了什么。人话就是——你的消融表里必须有一行，是打不赢朴素 RAG 的复杂度。\n\n仓库 README 致谢 GoodMem 团队的独立复现：同样的 embedding 模型、reranker 和检索深度跑 LoCoMo，GPT-4o-mini 读手得 88.68、GPT-4.1-mini 得 93.48，论文报告 89.07 和 93.70，四个数都落在 95% 置信区间内。\n\n## 所以呢\n\n这篇论文的价值可能不在「朴素 RAG 赢了」，而在把一个长期没人愿意验证的前提摆上台面：智能体记忆的复杂度，到底是问题带来的，还是行业共识带来的。\n\n它测的全是「给定一段已存在的历史，如何回答问题」。真实产品的记忆问题还有另一半——什么时候该记、记什么、什么时候该忘、数据主体要求删除时怎么办。这恰恰是结构化记忆系统在卖的东西。\n\n更准确的读法不是「记忆系统没用」，而是：证明自己比「把原始对话扔进向量库加重排」更强之前，复杂的记忆架构很可能只是在为想象出来的需求买单。","https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.30755","7437aeb9-930c-4866-a2e9-48003c1a792b",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"6ad31a14-c0da-42df-81fd-564281f768db","agentic-ai",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"5e628969-6d2a-437f-998a-104e4b16cfb1","ai-progress",{"id":19,"name":20,"slug":20,"description":14,"color":14},"120fa59a-ff6f-4537-9bf5-f818df636a0e","benchmark",{"id":22,"name":23,"slug":23,"description":14,"color":14},"01598627-1ea6-4b27-a5d8-874971571a71","llm",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"b728d68a-683c-403a-a96c-516d34a4427b","en","Naive RAG Wins 6 of 7 Settings Against 12 Memory Systems","A new arXiv paper pits naive RAG against 12 agentic memory systems and wins 6 of 7 benchmark settings, arguing raw evidence beats structured memory.","For the past two years, agentic memory has been one of the hottest directions in AI engineering. Every framework races on the same thing: pull the information out of long conversation histories, organize it, index it, store it as structured memory, and let the agent retrieve it on demand. Mem0, MemMachine, Memora, EverMemOS — the names sound more and more like products, and the architecture diagrams get more and more elaborate.\n\nBut a paper posted to arXiv on October 22 invited all of this into a test track, ran it around, and reached a conclusion that stings: the most naive RAG setup won almost everywhere.\n\nThe paper is \"Is Complex Agentic Memory Always Needed? Naive RAG Is Stronger Than We Thought\" ([arXiv:2610.30755](https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.30755)), by Zhengye Bao, Shengtao Zhang, Haokai Zhang, Shiqi Lou, Muning Wen, and Minnan Luo at Xi'an Jiaotong University, 62 pages in total. The authors named their baseline Raw-RAG: no extraction, no structural organization, no LLM rewriting of history — just serialize the raw conversation with timestamps, chunk it, run dense retrieval plus reranking, and hand the evidence straight to a reader model.\n\n## How much it won\n\nEvaluation covers four benchmarks: LoCoMo (10 conversations, 1,540 non-adversarial questions), LongMemEval (500 examples), MEME (100 episodes, 1,188 phase-specific questions), and MemoryAgentBench (14 sources, 2,071 questions). The comparison set spans 12 memory systems on the LoCoMo and LongMemEval side and 7 on the MEME and MemoryAgentBench side, including Mem0, A-MEM, MemMachine, Memora, EverMemOS, EMem, SimpleMem, MemOS, MemU, Nemori, RecMem, and MemoryOS.\n\nThe result: across seven benchmark-model settings, Raw-RAG takes six of them, losing only on MEME with a DeepSeek-V4-Flash reader, where MemMachine wins. For the exact margins, the [official GitHub repository](https:\u002F\u002Fgithub.com\u002FLitmeb\u002FRAW-RAG) publishes the comparison table: on LoCoMo, 89.07 with a GPT-4o-mini reader against MemMachine at 88.12, and 93.70 with a GPT-4.1-mini reader against EverMemOS at 93.05; on LongMemEval, 87.53 against Memora at 87.40; on MemoryAgentBench the gap widens to 48.32 against Memora at 43.84 with GPT-4o-mini, and 53.41 against 47.88 with DeepSeek-V4-Flash. The abstract puts the range as leading the strongest compared system by 0.13 to 5.53 points.\n\nWhat is notable is that the LongMemEval margin is only 0.13 points — essentially a photo finish. The paper itself does not dress this up as a rout.\n\n## The sharper result is the counter-evidence\n\nIf it were merely \"the simple method can also tie\", this would be one more parity paper. What actually hurts is the three experiments that follow.\n\nThe first audits Mem0 write decisions. Across 33,902 write decisions on LoCoMo, UPDATE and DELETE together account for only 4.39%; across 428,858 write decisions on LongMemEval, only 2.46%. Effective mutations — ones that actually changed text or deleted content — come to just 1.90% and 1.34%. The authors then swapped UPDATE for append-only ADD and DELETE for NONE, and LoCoMo accuracy went up, from 67.36% to 69.42%. In this configuration, explicit memory rewriting bought no accuracy at all.\n\nThe second adds memory machinery to Raw-RAG. All three variants get worse in the same MemoryAgentBench setup: atomic fact extraction drops to 28.70 (19.62 below the 48.32 baseline), summary augmentation loses 5.81 points, and graph expansion loses 5.78. The paper explains this by noting that atomic fact extraction takes test-time learning from 47.68% down to 15.37%, because the structure of usable demonstrations is broken apart.\n\nThe third is cost. In the paper latency table, Raw-RAG finishes ingestion plus retrieval in 305.91 seconds on the same 10-episode, 110-question MEME slice; EMem takes 1,561.97 seconds, MemOS 2,769.56, MemMachine 3,515.74, Mem0 4,142.95, and EverMemOS 9,301.48. The paper states this is 5.1 times faster than the next system and 30.4 times faster than EverMemOS. And adding those three memory-management operations costs 21.6 to 109.3 million LLM tokens and 2.9 to 8.8 hours, where Raw-RAG needs 12 minutes for the same work.\n\n## Why\n\nThe mechanism the paper offers rests on two metrics: coverage and utilization. Coverage is whether the retrieved evidence is sufficient to answer the question; utilization is whether the answer comes out right once the evidence is there. On MemoryAgentBench, coverage spans 19.07% to 45.29% across systems, while utilization spans only 80.00% to 89.71%. The conclusion is that what separates systems is not how well they read evidence once they have it, but whether the needed evidence is supplied at all.\n\nThe paper also introduces path coverage to explain long-chain questions: on MEME tracking and cascade items, Raw-RAG reaches 62.34% conditional path coverage, MemMachine 56.70%, and MemU only 9.62%. Retrieving the answer value is one thing; also preserving the temporal relations and dependency chain needed to derive it is another, and that is exactly where memory systems lose points.\n\nThe sharpest blade is the complete-store comparison. The authors run three conditions: each system native retrieval, a complete-store oracle that uses ground truth to select all phase-visible memories, and a raw-evidence control that also uses ground truth but picks the original conversation segments. The result is that Raw-RAG and MemMachine, both of which retain raw conversational content in their evaluated stores, recover to 57.54% and 57.46%, while the raw evidence itself scores 61.60%. Even granting full access to processed memory stores does not close the gap to raw evidence. With the packing ceiling removed, all seven systems still sit 7.58 to 29.38 points below raw evidence, and every paired 95% confidence interval excludes zero.\n\nThe paper states it plainly: information dropped or distorted at ingestion time cannot be recovered through retrieval.\n\n## Who should worry\n\nThis is not a paper saying memory systems are useless, and the wording is carefully restrained. It draws three boundaries: the evidence holds only for offline conversation tasks on public benchmarks and cannot be extrapolated to interactive, multimodal, or real production settings; raw storage grows linearly with history length and raises storage and privacy concerns; and state reconstruction quality still depends on reader model capability. In the appendix, on PersonaMem-v2, Raw-RAG scores 49.62%, ranking third out of seven behind MemOS and EverMemOS, and leading no individual scenario.\n\nBut the acceptance bar it proposes is concrete. The paper says, in its own words, that improved memory systems should demonstrate gains over a well-specified naive RAG baseline and establish what the added complexity contributes. Translated into plain language: if you are building a memory system, the next benchmark table must contain a row for the complexity you cannot beat with naive RAG.\n\nThe community response is worth noting. The repository README carries a dedicated section thanking the GoodMem team for an independent reproduction: running LoCoMo with the same embedding model, reranker, and retrieval depth, they got 88.68 with a GPT-4o-mini reader and 93.48 with a GPT-4.1-mini reader, against the paper reported 89.07 and 93.70, both inside the 95% confidence intervals. That reproduction also shows up in the [comments on the HuggingFace paper page](https:\u002F\u002Fhuggingface.co\u002Fpapers\u002F2610.30755).\n\n## So what\n\nThe value of this paper may not lie in \"naive RAG won\" but in putting a long-unexamined premise on the table: is the complexity of agentic memory actually demanded by the problem, or by industry consensus?\n\nThe authors are fairly measured about it. They are not saying you should not build memory systems. They are saying that if you do, your ablations have to include a naive RAG baseline, and you have to report end-to-end cost — ingestion plus retrieval — not just accuracy. The gap between EverMemOS at 9,301 seconds and Raw-RAG at 305 seconds says more about the price of an architectural choice than any benchmark score does.\n\nOne thing the paper does not prove is worth flagging: everything it tests is \"given an existing history, how do you answer questions\". The other half of the memory problem in real products is when to remember, what to remember, when to forget, and what to do when a data subject asks for deletion. Those are exactly the things structured memory systems are selling, and exactly the things naive RAG cannot help with. The limitations section does touch on privacy and deletion constraints, just without going deep.\n\nSo the more accurate reading is not \"memory systems are useless\", but rather: until you prove you are stronger than \"throw the raw conversation into a vector database with a reranker\", a complex memory architecture is probably paying for requirements you imagined.","naive-rag-beats-complex-agentic-memory","2026-10-25T07:17:45Z","2026-10-25T07:18:10.789265Z","2026-10-25T07:18:10.789279Z",true,"agent",8,[39,48],{"slug":40,"tag_slug":40,"title_zh":41,"title_en":42,"intro_zh":43,"intro_en":44,"id":45,"is_active":35,"created_at":46,"modified_at":47},"ai-for-science","AI for Science 2026：从 UniPert 到 GPT-Rosalind 的硬核进化","AI for Science 2026: from UniPert to GPT-Rosalind","生命科学、化学材料、物理世界模型——AI 正在从\"语言工具\"变成\"实验伙伴\"。本专题收录 AI 在三大科学方向的关键节点：UniPert 统一基因与化学扰动空间、GPT-Rosalind 端到端生命科学推理、达摩院 AI 智能体 28 小时找到 4 种超导新材料、Anthropic Claude Science 把工作台做成标准品。","From language tool to lab partner — AI is reshaping life sciences, chemistry\u002Fmaterials, and physical world models. This topic covers the key milestones: UniPert unifying genetic-chemical perturbation spaces, GPT-Rosalind's end-to-end life-sciences reasoning, DAMO's AI agent discovering 4 superconducting materials in 28 hours, and Anthropic's Claude Science workbench going mainstream.","988a4300-5fab-41c4-b5d8-63711a2dc757","2026-09-10T01:34:15.296649Z","2026-09-10T01:34:15.296663Z",{"slug":49,"tag_slug":49,"title_zh":50,"title_en":51,"intro_zh":52,"intro_en":53,"id":54,"is_active":35,"created_at":55,"modified_at":56},"h3-series","MiniMax H3 系列：从开源权重到 35 倍吞吐","MiniMax H3 Series: from open weights to 35x throughput","MiniMax H3 自 2026 年 8 月开源以来节奏密集：官方把生成、参考与编辑收回一个模型；ComfyUI 当天压进 RTX 3060；摩尔线程 3 小时完成国产 GPU 适配；fal 后训练版把吞吐拉到 35 倍；FastH3 蒸馏再砍推理成本。本专题持续追踪 H3 的发布—开源—蒸馏—部署全链路。","Since MiniMax open-sourced H3 in August 2026 the pace has been relentless: one unified omni-modal model, same-day ComfyUI support down to an RTX 3060, a 3-hour Day-0 port to Moore Threads GPUs, fal's post-trained H3 Max at 35x throughput, and FastH3 distillation cutting inference cost further. This topic tracks the full H3 chain — release, open weights, distillation, deployment.","83ef0daa-3c31-4cb3-86ed-e5ee58654d5f","2026-09-08T07:33:19.942193Z","2026-09-08T07:33:19.942209Z",{"items":58},[59,64,69,74,79,84],{"id":60,"title":61,"news_slug":62,"published_at":63},"da6cbb98-3b89-4631-8dbd-59411136fb30","哈佛用 700 家公司拆解 AI 编程智能体:30% 代码增量,49% 合并时延","harvard-coding-agent-review-bottleneck","2026-10-14T11:30:00+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"c368ad9f-9308-4a0b-8f5c-3ae4601b48b9","D2K-Bench: 专家设计把 LLM 写 GPU 核提速 33.9%","d2k-bench-llm-gpu-kernel-design-guidance","2026-10-07T03:00:00+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"85f7f1c2-d896-436b-a915-37faed8776eb","你改主意了,模型没改:被拒需求也会带偏大模型","intent-eval-rejected-change-confusion","2026-10-06T17:15:00+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"206ea36a-1eea-463f-a241-1e3b32f5ec2d","USTC GraphForge:证据图把任务和 rubric 钉在一起,Qwen3.6-27B 涨三基准","graphforge-ustc-qwen36-27b-evidence-graph","2026-10-04T03:05:00+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"4c4e444f-9614-42fe-b8f4-743204f854fa","RL 后训练收「锐化税」:base 模型配轻 harness,pass@K 反超官方版","sharpening-tax-rl-post-training","2026-10-03T15:09:45+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"2035a9c7-2bc8-404d-9646-1813cbe4fa30","腾讯混元 LHTB：长程终端 Agent 最强仅 15.2% pass@1","tencent-hunyuan-lhtb-benchmark","2026-07-15T00:00:00+00:00"]