AI
AI
资讯
alishangtian.com
首页
大模型
智能体
开源项目
研究前沿
行业动态
知识库
主题色 · THEME
靛蓝(默认)
极光
落日
薰衣草
海洋
森林
暮橙
石墨
自定义
恢复默认
提交线索
agent
anthropic
openai
huggingface daily papers
ai安全
gpu
jake wharton
it之家
solidot
港股
搜索:
case
共命中 50 条(服务端检索)
Some Anthropic veterans are reportedly buying remote land in
case
"AI goes awry"
Sep 27, 2026 Some of Anthropic's longest-serving employees are making concrete contingency plans in
case
AI spirals out …
行业动态
The Decoder · 9-27
阅读 42
·
访客 42
The climate
case
for AI has not been made
This week, as countries meet in Antalya, Turkey for the United Nations climate conference, one topic likely to dominate …
行业动态
MIT Technology Review · 11-9
阅读 17
·
访客 17
EDGEGEN: Improving Tool-Calling Agents Beyond Happy Paths with Synthetic Edge
Case
Generation
Tool-calling LLM agents are increasingly deployed in enterprise applications. However, effective evaluation and optimiza…
智能体
HuggingFace Daily Papers · 9-21
阅读 75
·
访客 75
AI 安全论证:把“系统安全”写成一条可审查的证据链
安全论证不是一句“模型安全”,而是针对具体系统、用途和运行环境,把安全主张、推理路径、证据、假设与反例组织成可复核链条;本文说明如何定边界、拆危险、判断证据强弱并随变更更新。
大模型
精选
· 原创 · 昨天
阅读 3
·
访客 3
智能体的可观测性生态:tracing、调试工具与生产监控
智能体的非确定性和多步执行让传统调试手段失效,本文梳理 tracing 标准化、主流观测工具生态、生产监控的关键指标体系,以及从坏
case
到回归评测集的闭环实践。
智能体
精选
· 原创 · 11-5
阅读 81
·
访客 78
RAG 的生产监控与调试:追踪检索质量的生产链路
RAG 上线后的调试比开发期更难:真实流量的多样性、检索质量的隐性退化、生成质量的持续漂移——生产监控是 RAG 运营的眼睛。本篇拆解 RAG 监控的四个层面(检索层、生成层、用户层、系统层)、调试的方法论,以及 bad
case
的管理流程。
一叶一世界
精选
· 原创 · 11-4
阅读 65
·
访客 65
Making the
case
for context-aware AI
Sponsored In partnership withCelonis Enterprise AI can generate recommendations, automate tasks, and help organizations …
行业动态
MIT Technology Review · 10-22
阅读 26
·
访客 26
Cheaper AI tokens are driving more demand, and that's Jensen Huang's best-
case
scenario
Oct 11, 2026 Jevons' paradox is Jensen's best friend.** Data from Ornn, Silicon Data, and Bloomberg (as of August 2026) …
行业动态
The Decoder · 10-11
阅读 140
·
访客 140
OpenAI says a misaligned model deliberately destroyed its own environment hoping for a fresh start with better data
Oct 10, 2026 OpenAI has a few new rogue agent stories.** In the first
case
(October 6), an AI evaluation model couldn't …
智能体
The Decoder · 10-10
阅读 74
·
访客 74
AI math breakthroughs have Ethereum researchers debating how fast wallet security could collapse
Manuel Uth Oct 8, 2026 Two leading Ethereum researchers warn that AI-assisted math could, in the worst
case
, break the s…
行业动态
The Decoder · 10-8
阅读 29
·
访客 29
儿童因经常将笔记本电脑放在腹部而出现烤肤症]
《BMJ
Case
Reports》报告了一起奇特的病例,一名儿童因腹部皮肤出现奇怪斑痕而送去急症。医生查找许久未发现病因,因此开了抗生素让他回家,叮嘱父母如果病情变化立即来复诊。两周后,这名儿童再次入院,他的病情出现恶化。皮肤斑块变大,颜…
行业动态
Solidot · 9-11
阅读 35
·
访客 33
MetroLLM-Bench: Evaluating Language Models as Transit Kiosk Runtimes
We introduce MetroLLM-Bench, a 955-
case
benchmark for testing language models as the policy layer of a transit kiosk. It…
研究前沿
HuggingFace Daily Papers · 9-9
阅读 54
·
访客 48
Privacy Failure in Split-LLM Training, The Returned Gradient Nullifies the Decoys
We present a systems-security
case
study of a two-node split-LLM training system whose privacy evaluation passed while l…
大模型
HuggingFace Daily Papers · 9-3
阅读 45
·
访客 44
NVIDIA DGX Spark Developer Kit Turns a Desk Into a Supercomputing Lab
At SC26, an annual supercomputing conference taking place this week in Chicago, NVIDIA is introducing the DGX Spark HPC …
行业动态
NVIDIA Blog · 昨天
阅读 8
·
访客 8
Building trustworthy AI by fencing it in, not predicting it
Sponsored Provided byUL Solutions In October this year, OpenAI disclosed that two of its AI agents had switched off the …
智能体
MIT Technology Review · 昨天
阅读 5
·
访客 5
California state AG sues OpenAI over rogue agent incidents
OpenAI is facing a new lawsuit from the California Attorney General’s office, alleging that the company’s training pract…
智能体
TechCrunch · 昨天
阅读 2
·
访客 2
Uber says hackers stole information on close to 150,000 driver’s licenses during data breach
Uber is notifying close to 150,000 drivers that their personal and driver’s license information was stolen during an Oct…
开源项目
TechCrunch · 昨天
阅读 6
·
访客 6
Hark Releases Handoff², a 2x Faster Computer-Use Agent for Hark Pro
!(https://www.gstatic.com/images/branding/googleg/1x/googleg_standard_color_128dp.png)Add as a preferredsource on Google…
智能体
MarkTechPost · 2天前
阅读 6
·
访客 6
Audible turns to AI to help users discover audiobooks
Amazon-owned audiobook service Audible is turning to AI to help listeners find their next book. The company on Monday la…
行业动态
TechCrunch · 2天前
阅读 1
·
访客 1
Anthropic says the key to agent memory is knowing what to forget, not how much to store
Nov 16, 2026 Anthropic Key Points Anthropic tested how Claude can store information in note files across sessions and re…
智能体
The Decoder · 2天前
阅读 2
·
访客 2
Building AI that can work across the enterprise
Sponsored In partnership withInfosys For years, enterprises have approached AI as a series of discrete experiments. Now,…
智能体
MIT Technology Review · 2天前
阅读 5
·
访客 5
Asian governments react to Nvidia CEO Jensen Huang’s exit
Jensen Huang’s departure from Nvidia has drawn a range of responses across Asia, with some governments moving to confirm…
行业动态
TechCrunch · 3天前
阅读 19
·
访客 19
Australia's central bank finds AI hitting young workers first, especially in software and customer service
Manuel Uth Nov 15, 2026 Nano Banana Pro prompted by THE DECODER AI has barely displaced any jobs in Australia so far, bu…
行业动态
The Decoder · 3天前
阅读 5
·
访客 5
Gemini 4 Carbon takes the MathArena crown but costs 15 times more than GPT-6.1 Sol
Nov 14, 2026 Google's Gemini 4 Carbon has become the first non-OpenAI model in a while to top the MathArena math benchma…
研究前沿
The Decoder · 3天前
阅读 16
·
访客 16
AI agent startup LangChain hits $3B valuation
Agent management startup LangChain has raised a $300 million Series C at a $3 billion valuation led by CapitalG, with pa…
智能体
TechCrunch · 4天前
阅读 12
·
访客 12
Stanford Research Open-Sources Memorilla, a Latent Memory Module for Frozen LLMs
!(https://www.gstatic.com/images/branding/googleg/1x/googleg_standard_color_128dp.png)Add as a preferredsource on Google…
开源项目
MarkTechPost · 4天前
阅读 17
·
访客 17
Do Underwater Image Quality Metrics Agree with Human Perception? A Benchmark with Human Choices, Simple Edits and Generative Drift
Underwater image enhancement is challenging to evaluate due to the lack of clean reference images, so researchers rely o…
研究前沿
HuggingFace Daily Papers · 5天前
阅读 1
·
访客 1
Meta FAIR Releases e-means: An Algorithm Beating k-means in 10x Fewer Epochs
!(https://www.gstatic.com/images/branding/googleg/1x/googleg_standard_color_128dp.png)Add as a preferredsource on Google…
行业动态
MarkTechPost · 11-11
阅读 37
·
访客 37
Lazare, veni foras: Reprogramming Dormant Neurons to Backdoor AI Agents
Microsoft researchers have discovered “Lazarus,” a backdoor technique that can reprogram dormant neurons in an AI model,…
智能体
Microsoft Research · 11-11
阅读 32
·
访客 32
Reddit is turning user conversations into AI-powered product reviews
Shortly after announcing a plan to transition its site to a logged-in experience, Reddit is debuting a feature that coul…
行业动态
TechCrunch · 11-10
阅读 12
·
访客 12
FraudBench: A Legal Evaluation of AI Deception
We introduce FraudBench, an evaluation of large language models' (LLMs) propensities to recommend, assist, or directly t…
大模型
HuggingFace Daily Papers · 11-9
阅读 12
·
访客 12
The Download: empty AI climate claims, and an Nvidia investigation
This is today's edition of* *The Download*,*our weekday newsletter that provides a daily dose of what's going on in the …
行业动态
MIT Technology Review · 11-9
阅读 12
·
访客 12
Korea Advances Physical AI Across Industries With NVIDIA
Tapping into the country’s strengths in manufacturing and engineering, NVIDIA is helping advance Korea’s physical AI eco…
行业动态
NVIDIA Blog · 11-9
阅读 11
·
访客 11
RAG 与领域适配:通用模型到领域专家的微调与检索适配路径
通用 RAG 进专业领域会系统性失效,本文给出诊断—适配—回归的完整路径:先定位失效层,再按"先便宜后昂贵"的原则选择文档改写、embedding 微调、RAFT 或领域预训练。
后端技术
精选
· 原创 · 11-9
阅读 68
·
访客 67
RAG 的用户反馈循环:thumbs up/down 怎么变成检索改进的信号
点赞点踩不是现成标注数据,直接拿来训练会踩进偏差与归因两个坑。本文给出反馈归因、置信度加权、评测集沉淀与训练负例挖掘的完整闭环设计。
后端技术
精选
· 原创 · 11-9
阅读 49
·
访客 48
一篇读懂 LLM 安全对齐:Red Teaming 与安全训练的工程实践
从 GCG 的梯度攻击到 PyRIT 的自动化红队,拆解安全对齐的攻击面测绘、安全数据构造、过度拒绝权衡与多层防御。
大模型
精选
· 原创 · 11-9
阅读 42
·
访客 42
知识图谱增强 RAG:实体链接与关系检索的工程实现
不谈从零构建知识图谱,聚焦企业已有 KG 资产的接入工程:词典先行的实体链接、置信度阈值与"不链接"选项、一跳到两跳封顶的子图抽取、三元组语言化与三通道融合。
后端技术
精选
· 原创 · 11-9
阅读 53
·
访客 53
一篇读懂解码策略:Beam Search、Contrastive Search 与约束解码的完整对比
从似然陷阱到退化惩罚再到语法约束,拆解三类高级解码策略的底层机制、计算代价与适用边界。
大模型
精选
· 原创 · 11-9
阅读 38
·
访客 38
Formally verifying agent-written full-stack web apps with PDD Kit
At a glance Formal methods for everyday software.** PDD Kit helps you turn your intent into a formal spec, then proves t…
智能体
Microsoft Research · 11-6
阅读 14
·
访客 14
Knowledge Graph-based Retrieval-Augmented Generation for Schema Matching
Traditional similarity-based schema matching methods fail to resolve semantic ambiguities and conflicts in complex, doma…
行业动态
HuggingFace Daily Papers · 11-6
阅读 11
·
访客 11
企业 AI 采购的 TCO 账:从 POC 陷阱到规模化投产
企业 AI 项目的真实死亡率惊人:大量 POC(概念验证)死在规模化的路上——不是技术不行,是 TCO(总拥有成本)的账没算清。把企业 AI 的成本结构、POC 陷阱与规模化清单讲清,这可能是企业数字化老兵最需要的一篇。
行业动态
精选
· 原创 · 11-5
阅读 45
·
访客 45
智能体的测试策略:从单元到端到端的验证体系
传统断言测试在非确定性的智能体上失灵,本文给出智能体版测试金字塔——确定性单元测试、组件级评测、端到端轨迹验证三层体系,以及评测集建设、Judge 校准与回归门禁的工程实践。
智能体
精选
· 原创 · 11-5
阅读 62
·
访客 61
CARTBench: Rubric-Based Evaluation of Shopping Advice for Complex and Ambiguous Requests with Timelines
Large language model (LLM) shopping assistants now recommend products in multi-turn conversations. However, reliable eva…
智能体
HuggingFace Daily Papers · 11-4
阅读 2
·
访客 2
FlyHash for LLM Agent Memory: An Empirical Validation and a New Application
Large language model (LLM) agents increasingly rely on external memory stores to retain information across conversations…
智能体
HuggingFace Daily Papers · 11-4
阅读 12
·
访客 12
Rapid LLM development is outrunning the research meant to explain it, study finds
Nov 3, 2026 GPT-Image-2.5 prompted by THE DECODER Stanford researchers Owen Queen and James Zou tested 50 widely cited c…
大模型
The Decoder · 11-4
阅读 37
·
访客 37
RAG 的未来趋势:从检索到智能知识工作
RAG 从「检索增强生成」正在演化为「智能知识工作」:从被动检索到主动知识管理、从文档问答到知识推理、从单一管线到多模态多源。本篇展望 RAG 的五个趋势方向,以及它们对「知识工作」的长期影响。
一叶一世界
精选
· 原创 · 11-4
阅读 71
·
访客 70
一篇读懂结构化数据 RAG:表格、数据库与 Text-to-SQL
企业最有价值的知识往往不在文档里,在数据库里。Text-to-SQL 让 LLM 把自然语言变成查询语句,但这条路远比「生成一段 SQL」复杂。本篇拆解结构化数据 RAG 的完整管线:Schema 理解、SQL 生成、执行验证与错误恢复,以及它与向量 RAG 的本质差异。
一叶一世界
精选
· 原创 · 11-4
阅读 65
·
访客 65
训推一体与资源错峰:集群共享的调度设计
白天推理高峰、训练占卡;训练希望整周独占、推理要有兜底容量——训推共存的集群需要一个调度设计。本篇拆解训推一体的资源矛盾、错峰调度的三种模式(抢占、分区、弹性分区)、以及「在线优先」的调度纪律。
大模型
精选
· 原创 · 11-4
阅读 50
·
访客 50
Agentic RAG:让检索长出规划、路由与自我修正
Agentic RAG 把检索从「固定管线」升级为「智能体的决策」:要不要检索、检索什么、从哪检索、结果够不够好——每一步都由 Agent 判断。本篇拆解它的核心模式(检索路由、查询规划、自我修正循环)、与传统 RAG 的分工边界、以及 Agentic RAG 的成本与延迟代价。
智能体
精选
· 原创 · 11-4
阅读 48
·
访客 48
RAG 的 UX 设计:检索结果如何呈现给用户
RAG 的技术再好,用户看不到检索过程、不理解引用关系、无法判断答案的可信度——体验就失败了。本篇拆解 RAG 的 UX 设计:引用的呈现方式、过程的透明度、以及「让用户信任并有效使用 RAG」的交互设计原则。
一叶一世界
精选
· 原创 · 11-4
阅读 64
·
访客 64