看榜单选 embedding 模型经常翻车，因为榜单是通用语料，你的数据不是。本文给出语言与领域的第一分水岭、维度的真实代价、可信评测的做法与工程参数对照表，五十条真实问题即可完成一次有依据的决策。

语义检索的第一步是把文本变成可比较的向量。本文从 one-hot 的困境讲到稠密向量的语义几何，手算一个余弦相似度的小例子，再用十几行代码演示从 encode 到 top-k 的完整流程，最后点出语义匹配最常见的两个坑。

Cohere has released Embed 5, a new embedding model family. It targets enterprise search, RAG, and agentic retrieval. The…

Perplexity Research and turbopuffer have released **pplx-embed-v2-context-9b-preview**, a contextual embedding model for…

In this report, we introduce Ovis-Embedding, a state-of-the-art omni-modal embedding family built on native integration …

Linkup Research has released SPARSEUP, an open-source sparse embedding model built on a 149M-parameter ModernBERT backbo…

We present Multimodal Flow, a fully continuous generative model of language and vision. Most unified multimodal models e…

The rapid progress of artificial intelligence is reshaping robotics and accelerating the adoption of learning-based appr…

RAG、推荐、去重的共同底层需求是「找最相似的 K 个向量」。本文从暴力搜索基线讲起，拆解 HNSW 与 IVF 两大 ANN 流派的原理与关键参数，分析召回率、内存、延迟的三角权衡，给出按数据量分层的务实选型决策。

Extending a text embedding model to new modalities typically degrades text retrieval quality, and existing omni-modal em…

In this tutorial, we work with **MSEB**, the Massive Sound Embedding Benchmark from Google Research, and approach it fro…

We present ENEAS, a unified, text-promptable method for instance tracking and semantic discovery. Text-promptable segmen…

文档不能出内网的团队也想要自己的 ChatPDF。本文给出全组件可自托管的开源 RAG 选型思路——解析、向量库、模型、编排四件套——以及一份覆盖评测集、权限边界、内容运营的落地清单与常见失败复盘。

纯向量检索抓不住型号、缩写这类精确术语。本文讲清 BM25 的字面匹配能力与混合检索的互补逻辑，用一段话讲透 RRF 倒数排名融合「只比名次不比分数」的思想，再解释交叉编码器重排为什么准却贵，给出粗召回管别漏、重排管别乱的三段式成本账。

「感觉变好了」不算数。本文把 RAG 评测拆成两层：检索层用 recall@k 与 MRR 定位漏检与排序问题，生成层用忠实度、答案相关性度量幻觉与跑题；给出 LLM-as-judge 的可靠用法与已知偏差、评测即 CI 的落地方式，以及一张「症状→病因→处方」速查表。

大模型的知识停在训练截止日，私有知识它更是从未见过。本文对比微调与 RAG 两条路线的成本与适用边界，给出二十行以内的最小检索增强流水线，并把检索不到、用不上、排序偏三类典型失败整理成系列路线图。

Agent 工程系统学习第 4 章：记忆系统解决跨会话问题。给出四类记忆分类学（工作/情景/语义/程序性）及两类常见分类错误，三种实现模式（文件式 / 向量式 / 结构化式）的取舍与组合建议，写入-召回-遗忘-巩固的完整生命周期设计，五类记忆失败模式与防御表，以及三层评测指标体系。

鹭羽* 2026-10-01 23:06:30 来源：量子位 用ImageNet训练encoder 克雷西 发自 凹非寺 量子位 | 公众号QbitAI 教会AI做ARC抽象推理题的，竟然是猫猫？ 何恺明团队最新论文提出了**NAT-AR…

Keyword-matching benchmarks can credit small models for tool use they never perform. We document such a false positive i…

What makes great scientists great? Even as AI systems start to make progress on open problems, scientists remain far ahe…

NVIDIA has released Kumo Tabular, a new family of tabular foundation models (TFMs) for classification and regression. If…

VectifyAI 开源的无向量 RAG 引擎 PageIndex（当日涨星 +1,095、★38,082、MIT）：把长文档编译成 JSON 层级树，让 LLM 逐节点推理导航，取代切块+向量相似度。基于它的 Mafin 2.5 在 FinanceBench 全量 10,231 题报告 98.7%。拆解两阶段架构、三模式 TOC 自校验回退、三工具检索循环、成本口径，以及多文档规模化与数据主权的真实边界。

NVIDIA联合超过100家行业伙伴推出开放的AI智能体安全平台，核心理念是安全控制不应运行在被控制的智能体内部：OpenShell作为Apache 2.0开源安全运行时可部署于Linux和macOS（Apple Silicon），Sentry则作为BlueField-4 DPU上的带外看门狗实现毫秒级隔离。

Lookup-based memory has been a promising way to scale the parameters of large language models (LLMs). It retrieves learn…

Listen on Apple Podcasts Listen on Spotify Meta’s new AI agent Muse took the spotlight at the company’s annual Connect e…

Sarvam AI has released Saaras V4, the newest generation of its speech recognition model. It covers all 22 scheduled Indi…

Liquid AI has announced LFM2.5-VL-3B-DSpark, an experimental speculative-decoding draft model for its LFM2.5-VL-3B visio…

田, 晏林* 2026-09-26 17:01:00 来源：量子位 GitHub现在最火热的大模型开源小蜂鸟Colibrì是个啥？ 闻乐 发自 凹非寺 量子位 | 公众号 QbitAI 25GB笔记本硬跑744B GLM-5.2，32GB…

In this tutorial, we build a comprehensive multimodal augmentation and robustness workflow with **AugLy** for images, te…

A main promise of looped language models is depth-adaptive inference. By looping a block of shared layers a variable num…

🧑‍💻GPT-6 Sol 与 Luna 发布，API 价格降低 50% 🧠Claude Opus 5.5 发布，典型任务运行成本降低 40% 🍎特努斯：Siri AI 不应代替人际关系 📱OPPO Find X10 系列发布：49…

Builder.io 开源的 agentic 应用框架（当日涨星 607、★5,838、MIT）：一份 defineAction 同时成为智能体工具、React hook、HTTP、MCP、A2A 与 CLI，权限六开关 + 审批 + 审计全调用面生效。拆解动作层架构、约 40 个停止条件的运行时与五个落地场景。

Degradations vary widely across images, so a practical restoration system has to handle many degradation types with one …

TypeSafe AI 的 Jev 全面开放，注册即得 5 美元额度（约 1.2 亿输入 Token），输出 Token 永久免费。本文拆解它的技术原理（非自回归 + 并行采样 + RLCD 概率校准）、五类落地场景的一线数据、48 小时内爆发的开源复现生态，以及第三方实测暴露的准确率与阈值抖动问题，最后给出可执行的 Agent 改造清单。

Standard positional encodings represent position as a one-dimensional reading-order coordinate, but reading order alone …

Despite impressive visual quality, state-of-the-art video diffusion models often generate content that violates real-wor…

同一预测核心，跨七类系统验证]

Standard subword tokenizers either treat every orthographic variation of a word (such as hello, Hello, HELLO, and Héllo)…

Multi-modal image generation, particularly subject-driven customization, has garnered growing attention in recent years.…

Anthropic and OpenAI want to embed independent safety evaluators inside their AI labs. Researchers welcome the unprecede…

GLiFormer Large scores 91.10 F1 on nested JSON, near GPT-5.6-luna's 91.96, while grounding every value in source spans. …

World modeling enables intelligence to anticipate consequences, guide interventions, and learn from interaction. Yet pre…

Google Research has introduced Retrieve-for-Train (R4T), a framework for search that returns coherent, diverse result se…

Autonomous research loops such as AutoResearch show that one coding agent can improve a training setup unattended. Run s…

把 AI 计算放在最适合计算 AI 的地方去。 #欢迎关注爱范儿官方微信公众号：爱范儿（微信号：ifanr），更多精彩内容第一时间为您奉上。 ]

系列第 ③ 篇，对应技能地图第 2 格「Grounding」。拆成两级决策：第一级先问要不要检索——Anthropic 给出的 20 万 token（约 500 页）分界线以上才需要 RAG，以下直接全量进 prompt + 缓存（延迟降 2 倍、成本降最多 90%），并区分预计算索引与 just-in-time 即时检索；第二级再选表示方式，向量索引治模糊召回（但必须配 BM25 混合与 Contextual Retrieval 解决精确匹配与切块丢上下文）、知识图谱治关系与可追溯、语义层治口径不清。附可量化收益表（检索失败率 5.7% → 3.7% → 2.9% → 1.9%）、四个实现注意项、context rot 与上下文压缩/笔记/子智能体三件套，以及一张可抄的选型决策树。

World-action models (WAMs) jointly model future observations and actions, typically predicting the future as RGB images.…

Scaling large language models efficiently has motivated sparse capacity mechanisms such as Mixture-of-Experts and, more …

We describe our entry to the EgoLongQA track of the Wearable-AI Challenge in ECCV 2026, which placed first in the

Reducing audio-encoder depth lowers the inference cost of speech large language models, but removing complete blocks per…