独立 AI 资讯库
从自动抓取数据库实时读取，适合快速浏览、检索和后续扩展成完整资讯站。
论文提出 FCPRAG（Fusion-Controller Parametric Retrieval-Augmented Generation），一种面向稳定多段落 LoRA 注入的参数化检索增强生成方法。参数化检索增强生成（PRAG）通过与段落对应的 LoRA 适配器，将检索到的证据注入大语言模型（LLM），从而减少对长上下文提示词的依赖。该研究聚焦于同一查询检索到多个段落时的融合与稳定注入问题。论文编号为 arXiv:2608.21750v1，发布类型为新论文。
论文提出 LëtzCross，这是一个用于评测卢森堡语文档跨语言、页面级多模态检索能力的基准。研究关注 ColPali 等页面图像检索器在视觉信息丰富文档上的表现，并指出其在跨语言、低资源语言环境中的行为仍缺乏了解。该基准可帮助开发者和研究人员评估多模态检索系统在低资源语言文档场景中的跨语言泛化能力。
论文《Mitigating Database Leakage in RAG Systems with Keyword-Grounded Fact Substitution》提出一种面向 RAG 系统数据库泄露风险的缓解方法。该研究关注提示注入攻击：攻击者可能误导检索器或影响系统对外部知识的使用，从而导致敏感数据库内容泄露。论文以关键词锚定的事实替换为核心思路，旨在提升 RAG 系统在使用外部知识源时的安全性。该工作以 arXiv:2608.21656v1 形式发布，对构建面向生产环境的 RAG、Agent 和知识库 API 具有安全参考意义。
随着 LLM 的进步以及 RAG 系统在行业中的应用，面向专有数据评估 RAG 性能的领域专用问答数据集需求正在上升。论文提出 TRIAD，用于自动化多跳 RAG 评估流程，覆盖从上下文提取到经验证数据集生成的环节。原文指出，HotpotQA 等现有数据集已对当前 RAG 系统构成挑战，但行业场景仍需要更贴合私有数据的评测集。该工作对开发者的意义在于，可能为构建和验证多跳检索增强生成评测数据集提供自动化方法。
这篇论文研究大语言模型（LLM）工具增强型 Agent 中的“遗忘”问题。作者指出，Agent 的回答不仅取决于模型参数，还会受到工具调用及外部观察结果的影响，因此传统 LLM 遗忘评估存在不匹配。论文聚焦于面向工具使用场景的遗忘方法与评估，对构建可控、可审计的 Agent 具有潜在意义。该论文为 arXiv:2608.21544v1，新投稿类型为 new。
这篇论文探讨了“Agentic Security”，即利用大语言模型（LLM）Agent 规划、调度和解读安全工具的渗透测试方法。作者指出，随着此类系统从演示原型走向实际部署，开发者和安全从业者反复遭遇相似的运营与执行失效问题。论文旨在系统梳理相关工具、常见失效模式及设计法则，为构建更可靠、可控的 LLM 驱动渗透测试系统提供参考。该研究对安全自动化产品、Agent 工作流设计及安全测试风险治理具有现实意义。
研究发现，现代多语言 tokenizer 对乌克兰语等代表性不足的西里尔字母语言切分更细，导致成本上升、可用上下文容量下降。在 BrUK 和 Brown 语料库上，乌克兰语相对英语的全文 token 开销在现代 tokenizer 中达到 68%—121%，较旧的 cl100k 中则达到 220%。研究测试的 LLMLingua-2 可在电商 RAG 基准中将乌克兰语输入长度减少 47%—49%，而采用均衡字节级 BPE 设计的 tokenizer 可将留出集的乌克兰语/英语 token 比从 2.22 倍降至 1.30 倍。结果表明，训练数据中的文字系统分配会影响分词效率，开发者可在推理压缩和 tokenizer 设计两个阶段进行缓解。
arXiv:2608.21377v1 Announce Type: new Abstract: Sycophancy in large language models, the tendency to prioritize user agreement over truthful responses, has been documented extensively but studied primarily in single-turn settings. This paper investigates a critical question: does subjecting LLMs to
arXiv:2608.21369v1 Announce Type: new Abstract: Nigerian Pidgin is one of Africa's most widely spoken languages, yet remains severely underrepresented in language model evaluation. Existing benchmarks primarily focus on translation, transcription, or generic sentiment analysis, leaving critical asp
arXiv:2608.21365v1 Announce Type: new Abstract: As a low-resource language, Khmer presents several retrieval challenges, including limited annotated data, ambiguous word boundaries, weak support in multilingual embedding models, and frequent mixed Khmer-English usage. This paper presents KSE-Web, a
arXiv:2608.21702v1 Announce Type: new Abstract: Retrieval-Augmented Generation (RAG) grounds LLM generation on retrieved documents, but the standard terminal retrieval stage--dense-vector similarity, optionally followed by reranking--often returns documents that share keywords with the query withou
arXiv:2608.21690v1 Announce Type: new Abstract: LLM agents increasingly take on long-running tasks whose history grows far beyond a single model context window. Existing approaches compress earlier interactions or extract selected information into fixed memory representations, committing to what to
arXiv:2608.21664v1 Announce Type: new Abstract: Safe deployment of increasingly capable models will likely come to rely on latent-space monitoring as a complement to behavioral evaluations, especially when evaluation-aware models exhibit scheming or deception. However, if models can also control th
arXiv:2608.21614v1 Announce Type: new Abstract: Chain-of-thought (CoT) prompting improves LLM reasoning by decomposing complex problems into intermediate steps, but its sequential nature increases decoding latency and memory usage. Mixture-of-Experts (MoE) models scale capacity through sparse exper
arXiv:2608.21610v1 Announce Type: new Abstract: Retrieval-augmented generation relies mostly on flat, fixed-granularity indexes: documents are cut into uniform chunks and retrieved by similarity, discarding the hierarchical structure of the source. We introduce Semantic Compression Trees (SCT), a h
arXiv:2608.21601v1 Announce Type: new Abstract: Benchmarks for scientific artificial intelligence are mostly written to be scored: multiple-choice questions, curated agent tasks with reference solutions, or simulators with a known generative structure. Real scientific requests arrive differently. T
arXiv:2608.21449v1 Announce Type: new Abstract: Time series arise in a wide range of application domains and are analyzed using machine learning in decision-critical settings. Time series classification (TSC) is one of the most widely studied and relevant tasks. In this context, ensuring the transp
arXiv:2608.21444v1 Announce Type: new Abstract: Multi-drone systems are increasingly positioned for safety-critical missions such as search and rescue (SAR) and critical infrastructure monitoring. Yet, real-world adoption remains constrained not only by autonomy performance, but by the difficulty o
arXiv:2608.21430v1 Announce Type: new Abstract: Multimodal language models increasingly show promise for enabling the large-scale computational analysis of film, opening up new avenues for learning about film history and the evolution of narrative techniques. But the creation of stable benchmarks b
arXiv:2608.21418v1 Announce Type: new Abstract: Manufacturing knowledge graphs that integrate data from heterogeneous industrial systems face a trust deficit: consumers cannot determine whether queried data is valid, whether it was valid when a decision was made, where it originated, or how it was
arXiv:2608.21417v1 Announce Type: new Abstract: Flexible manufacturing requires industrial robots to be reprogrammed rapidly as product variants change. This paper presents a language-model-based workflow that generates, validates, and iteratively corrects ABB RAPID robot programs from natural lang
arXiv:2608.21412v1 Announce Type: new Abstract: Root cause analysis (RCA) in large datacenter networks is challenging because telemetry is noisy, partial, and asynchronous. Score-based approaches degrade under these conditions, often yielding unstable or incorrect attributions. We present \textsc
arXiv:2608.21393v1 Announce Type: new Abstract: Running large language models inside enterprise environments has always bumped up against a practical wall: the data lives in one place, the AI horsepower sits somewhere else, and moving sensitive records between the two creates real headaches around
arXiv:2608.21375v1 Announce Type: new Abstract: Heterogeneous agentic retrieval-augmented generation (RAG) systems increasingly orchestrate external APIs, internal databases, vector stores, and graph stores. Exposing all tool descriptions to an LLM agent, or selecting tools only by vector similarit
arXiv:2608.21374v1 Announce Type: new Abstract: Literature reviews are essential to scientific progress, but rigorously evaluating automatically generated reviews remains difficult because many aspects of research utility depend on expert judgment rather than reference-overlap metrics. We introduce
arXiv:2608.21372v1 Announce Type: new Abstract: This report summarizes the work conducted on the Game of Hidden Rules (GOHR), focusing on reinforcement learning agents trained to infer hidden rules from trial-and-error feedback, representation design, rule difficulty analysis, transfer learning, ge
给 13 套工作流后端做同一个 App 检查更新接口，没建一张表：版本号的真相交给分发平台，任何一环失败都只说没有更新。从发版到装机一共九道，附真机跑通的一轮自我升级。
项目介绍 JimuChatBI 积木问数（Chat2BI）是积木报表推出的对话式智能数据分析产品，也是目前国内首款免费开源的 Chat2BI 智能问数方案 :sparkles: 它以语义建模构建可信数据底座，业务同学无需写 SQL、不懂建模，用自然语言提问即可秒级拿到数据表格、可视化图表和背后的 SQL，把原本需要排期两三天的取数流程，变成一场...
一场两小时的客户拜访，双方说的话加起来上万字。但回到工位，销售能凭记忆敲进CRM的，通常不到三百字。剩下的去哪了？跟着出租车一起走了。这样的对话在企业里每天都在发生。会议室里的争论、客户办公室里随口的一句提醒、老师傅肚里的经验，信息密度最高的这些话，几乎都发生在电脑屏幕之外，从来没有进入过任何一套企业系统。另一边，已经沉淀在文档、群聊和系统里的数据，也常常因为没做结构化，或者体量太大，塞不进上下文窗口，Agent根本调用不了。9月22日云栖大会，千问办公把两件事放到了一起解决：发布了面向企业级Agent的基础设施“企业上下文”（Enterprise Context），以及Agent硬件Qwen
很多人尝试给 Agent 接数据库，第一反应往往是拿本地脚本直连，或者找个现成的数据库 MCP 暴露 DDL 和执行接口。Demo 很容易跑通，但一旦进真实场景就会发现：大模型搞不懂缩写和数字枚举、复杂多表联查极易写错、生产账号密码暴露在客户端、甚至一条没有 WHERE 的 SQL 就能引发越权或误改。 要让 Agent 真正安全、稳定地...