T
traeai
Sign in

概念

什么是 RAG?

也叫:Retrieval-Augmented Generation

检索增强生成技术

为什么现在值得关注?

最近变化

2026-10-20 · 持续预训练使模型领域知识提升3倍(C3 AI案例中从11%到60%)

RAG 被反复提及时,通常意味着它正在影响产品路线、开发者工作流或 AI 产业判断。这个页面把分散材料合并成一个可持续更新的观察入口。

📰 RAG 最新动态

已收录 30 篇与「RAG」相关的 AI 资讯和分析。

RAG Is Burning Money — I Built a Cost Control Layer to Fix It

RAG Is Burning Money — I Built a Cost Control Layer to Fix It

Towards Data Science4995 字 (约 20 分钟)
92

RAG systems often incur hidden costs due to context over-fetching, lack of caching, and no model routing; the author built a cost control layer using semantic caching (98.5% hit rate), query routing (81% requests shifted to low-cost models), and token-budget circuit breaking, achieving 85.8% cost reduction at 10k requests/day without quality loss.

入选理由:上下文过取使每查询平均多消耗350 tokens,10k请求/日造成$52.5/日浪费(按$0.015/1K tokens计)

FeaturedArticle#RAG#Cost Optimization#Semantic Caching#Model Routing#LLM英文
Enterprise Document Intelligence: A Series on Building RAG Brick by Brick, from Minimal to Corpus scale

Enterprise RAG systems should focus on document understanding and business logic rather than stacking models and frameworks. Simple Python scripts often outperform complex production systems.

入选理由:多数企业RAG部署效果不佳,因基础解析和检索质量差。

FeaturedArticle#RAG#Enterprise AI#Document Intelligence#Retrieval-Augmented Generation#LLM Applications英文
Most people use vector databases for chatbots and RAG pipelines. 𝗦𝗲𝗻𝗾𝗶 𝗔𝗜 𝘂𝘀𝗲𝘀 ...

Senqi AI 使用 Milvus 向物理机器人注入长期语义记忆能力,解决真实世界任务中环境动态、任务无界、指令模糊和错误高成本等核心挑战。

入选理由:物理机器人Agent需实时重规划,因环境持续变化且任务无明确终点

FeaturedTweet#Milvus#RAG#机器人#向量数据库#AI Agent中文
From Regex to Vision Models: Which RAG Technique Fits Which Problem

From Regex to Vision Models: Which RAG Technique Fits Which Problem

Towards Data Science4997 字 (约 20 分钟)
90

RAG techniques are not universal; choose based on document structure and query control: use regex for templated docs, LLMs for sarcasm detection in transcripts, and vision models for schematics.

入选理由:模板化文档(如保险单、银行流水)适合用正则表达式提取字段,避免使用高成本的 RAG 流程。

FeaturedArticle#RAG#LLM#Document Intelligence#Vision Models#Enterprise AI英文
RAG Is Not Machine Learning, and the ML Toolkit Solves the Wrong Problem

RAG Is Not Machine Learning, and the ML Toolkit Solves the Wrong Problem

Towards Data Science6346 字 (约 26 分钟)
87

RAG is not machine learning, and the ML toolkit solves the wrong problem. The article argues that despite its resemblance to ML, RAG is fundamentally a search system, not a model, making hyperparameter tuning and embedding fine-tuning ineffective and misleading.

入选理由:RAG 解决的是确定性答案查找问题,而非预测未知结果,因此不能用 ML 方法优化。

FeaturedArticle#RAG#Machine Learning#Enterprise AI#Information Retrieval#LLM英文
Towards Data Science 图标

Proxy-Pointer RAG: Solving Entity and Relationship Sprawl in Large Knowledge Graphs

Towards Data Science3847 字 (约 16 分钟)
87

Proxy-Pointer RAG reduces the computational cost of entity and relationship reconciliation in knowledge graphs by over 90% by preserving document structure, enabling millisecond-scale ingestion without full-graph traversal.

入选理由:Proxy-Pointer RAG 使用 Skeleton Tree 和 Breadcrumb Injection 技术,使向量检索能精准定位文档完整结构段,而非碎片化块。

FeaturedArticle#RAG#Knowledge Graph#Proxy-Pointer#Entity Resolution#Vector Retrieval英文
𝗬𝗼𝘂𝗿 𝗥𝗔𝗚 𝘀𝘆𝘀𝘁𝗲𝗺 𝗽𝗿𝗼𝗱𝘂𝗰𝗲𝘀 "𝗵𝗶𝗴𝗵𝗲𝗿-𝗳𝗹𝘂𝗲𝗻𝗰𝘆 ...

Your RAG System Produces 'Higher-Fluency Hallucinations'

Weaviate • vector database(@weaviate_io)245 字 (约 1 分钟)
87

Research reveals poor retrieval quality is the primary cause of high-fluency hallucinations in RAG systems—more convincing, confident, and wrong—while scaling models fails to fix the root issue.

入选理由:检索质量差是RAG输出退化的最主要预测指标,模型能力增强反而加剧幻觉可信度。

FeaturedTweet#RAG#Vector Database#Weaviate#LLM#Hallucination Detection中英混合
What I learned from companies training their own models

What I learned from companies training their own models

Gradient Flow1367 字 (约 6 分钟)
85

企业自研模型时,持续预训练和模型工厂构建比单一模型优化更重要。多家企业通过混合训练策略实现领域知识深度适配。

入选理由:持续预训练使模型领域知识提升3倍(C3 AI案例中从11%到60%)

FeaturedArticle#模型训练#持续预训练#企业AI#模型工厂英文
AI HOT 精选 图标

DeepSeek 开放自研搜索 API

AI HOT 精选41 字 (约 1 分钟)
85

DeepSeek 推出自研搜索 API,支持多语言及 SEO 优化,日均处理数亿次查询。

入选理由:DeepSeek 搜索 API 日均处理数亿次查询,支持多语言检索

FeaturedArticle#DeepSeek#搜索API#AI#SEO优化中英混合
Towards Data Science 图标

How Do Word Embeddings Capture Relationships Between Words?

Towards Data Science3589 字 (约 15 分钟)
85

词嵌入通过向量空间中的几何关系捕捉词语间的语义关联,是NLP中实现语义理解的关键技术。

入选理由:词嵌入将词语映射为向量,通过向量距离反映语义相似性

FeaturedArticle#NLP#词嵌入#机器学习#AI应用英文
从 RAG 失败到 91% 准确率:Uber 法务红线 Agent 的四次架构演进

Uber 每年谈判数千份合同,法务审核是交易、供应商入驻、产品上线的关键瓶颈。团队的核心洞察是:这份看似高度专业的...

Uber通过四次架构演进,将法务红线Agent准确率提升至91%,合同审核时间下降20%+,实现AI与律师协同的法律审核系统。

入选理由:第二版引入指数衰减加权(半衰期365天)对抗策略漂移

FeaturedTweet#RAG#AI Agent#法律科技#合同审核中英混合
Rethinking access control for RAG with Amazon Quick and Amazon Bedrock

Rethinking access control for RAG with Amazon Quick and Amazon Bedrock

AWS Machine Learning Blog1181 字 (约 5 分钟)
85

AWS提出通过实时ACL验证解决RAG访问控制难题,确保AI生成内容符合企业权限策略。

入选理由:现有RAG方案依赖定期同步的ACL,存在权限过期风险

FeaturedArticle#RAG#访问控制#Amazon Bedrock#AI安全英文
Machine Learning Mastery 图标

Synchronous vs. Asynchronous Agent Execution: Architecture Patterns for Production

Machine Learning Mastery2042 字 (约 9 分钟)
85

同步和异步代理执行模式各有优劣,选择需基于任务复杂度、延迟容忍度和基础设施需求,异步模式更适合长流程任务。

入选理由:同步模式受AWS API Gateway 29秒超时限制,复杂任务易导致连接中断。

FeaturedArticle#AI#架构设计#LLM代理#同步/异步执行英文
EP227: Top 9 Places to Use Jev

EP227: Top 9 Places to Use Jev

ByteByteGo Newsletter1109 字 (约 5 分钟)
85

Jev模型凭借100倍于前沿LLM的性能优势,适合处理决策类任务而非生成任务,适用于模型路由、安全防护等9大场景。

入选理由:Jev模型性能是前沿LLM的100倍,适合处理实时决策任务

FeaturedArticle#Jev#LLM#AI Agent#系统设计中英混合
Building a RAG Pipeline for Semantic Code Search: A Developer Diary and Field Notes

Building a RAG Pipeline for Semantic Code Search: A Developer Diary and Field Notes

The JetBrains Blog3917 字 (约 16 分钟)
85

JetBrains分享了构建RAG管道进行语义代码搜索的实践经验,涵盖解析、分块、向量化等关键步骤,揭示生产环境中的技术挑战与优化策略。

入选理由:代码解析与分块是RAG管道的基础,需处理多语言语法差异

FeaturedArticle#RAG#语义搜索#代码库#JetBrains#AI英文
KDnuggets 图标

下一代智能企业系统需结合混合检索、多代理系统与自适应学习,以解决RAG在复杂场景下的局限性。

入选理由:混合检索结合向量搜索和BM25,解决企业场景下的术语差异和精确匹配问题。

FeaturedArticle#RAG#Agentic AI#企业AI#混合检索#多代理系统英文
Gallup scales real-time coaching for thousands with Amazon Bedrock

Gallup scales real-time coaching for thousands with Amazon Bedrock

AWS Architecture Blog1335 字 (约 6 分钟)
85

Gallup利用Amazon Bedrock构建AI助手,实现实时个性化职场指导,整合Claude模型与RAG技术,7天完成生产级部署。

入选理由:Amazon Bedrock Knowledge Bases实现RAG,确保回答基于验证研究

FeaturedArticle#Amazon Bedrock#RAG#AI助手#AWS架构英文
Why Your RAG System Is Only as Good as Its Translator Model

Why Your RAG System Is Only as Good as Its Translator Model

ByteByteGo Newsletter3034 字 (约 13 分钟)
85

RAG系统的性能高度依赖嵌入模型质量,选择合适的嵌入模型是构建有效RAG系统的关键。

入选理由:嵌入模型决定RAG系统搜索准确性,错误嵌入会导致错误答案

FeaturedArticle#RAG#嵌入模型#AI#信息检索英文
Modernizing and scaling support operations with generative AI on AWS

Modernizing and scaling support operations with generative AI on AWS

AWS Machine Learning Blog4067 字 (约 17 分钟)
85

AWS通过生成式AI重构支持运营,实现SOP自动化与SLA风险预测,提升效率30%以上。

入选理由:RAG技术使票务处理效率提升40%,缩短问题解决时间50%

FeaturedArticle#AWS#生成式AI#RAG#机器学习#支持运营英文
How to extract meaning from charts and tables in PDFs

How to extract meaning from charts and tables in PDFs

Weaviate Blog2871 字 (约 12 分钟)
85

Weaviate提出Late Interaction RAG方法,通过多向量模型直接解析PDF图表,解决传统RAG无法提取图表信息的缺陷。

入选理由:传统RAG处理PDF图表时,30%的查询会因图表信息缺失导致错误

FeaturedArticle#RAG#PDF处理#Weaviate#OCR#多向量模型英文
Introducing the MLPerf End-to-End RAG Inference Benchmark

Introducing the MLPerf End-to-End RAG Inference Benchmark

MLCommons2766 字 (约 12 分钟)
85

MLPerf推出首个端到端RAG推理基准,覆盖向量数据库构建与多跳问答流程,揭示多模型协作优化空间。

入选理由:RAG系统需多模型协作,单模型基准无法衡量其迭代推理行为

FeaturedArticle#MLPerf#RAG#AI推理#基准测试英文
Towards Data Science 图标

8 Tips for Writing Effective Agent Instructions

Towards Data Science1945 字 (约 8 分钟)
85

编写有效代理指令的8个技巧,涵盖流程图设计、明确目标、工具整合等关键点。

入选理由:使用流程图工具(如Mermaid.js)可视化业务流程可提升团队协作效率

FeaturedArticle#AI代理#指令工程#LLM#RAG#流程设计英文
Towards Data Science 图标

Why RAG Complexity Should Be Earned

Towards Data Science4406 字 (约 18 分钟)
85

RAG架构的复杂性应在检索失败模式被验证后逐步引入,而非默认采用。传统检索方法在特定场景下表现优异,过度复杂化可能掩盖基础问题。

入选理由:优先评估基础检索子系统,再引入复杂技术如重排序或代理系统。

FeaturedArticle#RAG#信息检索#生成模型#AI架构英文
Towards Data Science 图标

FAQ as RAG: When You Get to Design the Corpus

Towards Data Science4560 字 (约 19 分钟)
85

将FAQ作为RAG知识库可显著优化生成成本并提升检索效率,通过结构化设计实现解析简化与缓存复用。

入选理由:FAQ结构使RAG解析步骤复杂度降低70%

FeaturedArticle#RAG#FAQ#企业文档智能#生成模型英文

与「RAG」经常一起出现的 AI 术语。

💡 想追踪「RAG」的长期趋势?去 实体雷达 · RAG 查看详细分析和跨材料问答。

AI may generate inaccurate information. Please verify important content.