Tag
A Wired guide on when AI chatbots like ChatGPT, Claude, and Gemini are reliable for web search versus when traditional search is safer, noting AI excels at basic lookups and complex multi-source queries but falters on obscure questions and can present confident answers from poor sources.
RustXNG, a SearXNG-based local search engine created by a teenage student and released as open source, delivers near-Exa search accuracy with full page scraping, no API keys, and ~3.3 seconds for 1000 results on a local server—potentially upending the assumption that AI web search requires paid APIs.
Reddit announced it is shifting to a primarily logged-in experience as AI-powered search dries up its Google-driven traffic, aiming to protect data from scraping and boost ad revenue through licensing deals. The company reported Q3 revenue up 53% to $895 million but warned U.S. daily active users could decline further.
Alphabet's Q3 2026 results showed revenue growth but a sharp stock drop after Google admitted its AI Mode search queries earn less revenue per query than traditional web search, prompting plans to increase ad load to offset the shortfall.
The author developed an educational visualization page that reveals the six core stages of AI search, from understanding user needs to displaying citations, helping users intuitively understand the principles of GEO (Generative Engine Optimization) and how AI selects content.
Google is expanding ads within its AI-powered Search experiences to international markets as core search ad growth slows to 12% YoY, with Alphabet reporting $127.6B in Q3 revenue and its stock falling 8% after results.
RadonBench is a new benchmark comparing classical and learning-guided search methods (exhaustive extension, greedy local search, PatternBoost, AlphaZero, AlphaEvolve) for constructing maximum-cardinality Radon-free lattice configurations. None of the audited runs reached the open targets of 11 points in Z^3 or 21 points in Z^4, but the study establishes reproducible near-miss baselines for AI-assisted mathematical discovery.
The Last Museum 是一个可对全球主要博物馆的 580 万件（更新数据约 630 万件）艺术品进行语义搜索的工具，让公众通过自然语言检索海量艺术藏品。
Exa 发布 3.0 版本，称其在全部八项基准测试中排名第一，并新增低价搜索类型 Exa Lite（每千次请求 4 美元），同时将 Dynamic Highlights 正式开放并大幅扩展索引数据。
Fudan University research (NDSS 2027) compared 17,211 citations from 10 Chinese and Western AI search engines and found sharply different source ecosystems — Qwen cited 96% Chinese-language sites even when half its questions were in English, with only ~20% source overlap between any two engines.
Grablo Spot 是一款免费的 AI NVR 软件，可将现有摄像头录制的视频按人员和车辆自动整理成卡片，并支持通过人脸、外观描述、车牌片段等进行跨摄像头检索，所有数据仅存储在本地设备。
Exa，一家为AI提供搜索服务的公司，年化收入突破1亿美元，连续三年每年增长10倍，每月处理20亿次搜索且几乎全部由AI发起。文章分析其增长逻辑：先以低价获取搜索量，再通过提升每次搜索的价值实现收入跃升。
A Fudan University team (accepted at NDSS 2027) showed that a $14 payment to a Chinese content vendor planted a fake research concept into Doubao's answers within 90 minutes, and that 8 of 10 major AI platforms—including ChatGPT, Google AI, Perplexity, and Grok—cited the fabricated concept within a week, revealing how easily AI search citations can be gamed and how much Chinese chatbots favor their parent companies' own domains.
文章解析AI搜索赛道的赢家：自建网络索引的Exa估值达22亿美元，Tavily被收购（SEC文件显示交易金额为1.89亿美元而非传闻的4亿美元），并指出AI搜索的默认入口正掌握在模型公司手中（如Claude编程工具文档称搜索后端不可替换）。
Paul Graham 引用转发 Exa 创始人 Will Bryk 的推文，惊叹 Exa 年营收突破 1 亿美元、连续三年每年增长约 10 倍（累计约 1000 倍），每月处理 20 亿次请求、服务 20 万+ 团队，并预测到 2030 年 AI agent 的搜索量将是人类的 1000 倍。
LangChain announces AI Search in LangSmith, allowing users to search through traces conversationally to find answers like user frustration or agent failures instead of manually reviewing hours of data.
用户转发并评论 LangChain 在 Interrupt 大会上宣布的 LangSmith AI Search 功能，认为用自然语言搜索 trace 能大幅节省团队排查 agent 故障的时间。
Perplexity的CEO Aravind Srinivas透露公司营收同比增长5倍，并将增长归功于AI智能体热潮——Perplexity早已具备智能体所需的网页搜索、浏览器和多模型切换三大能力。
LangChain announces that p0 is sponsoring Interrupt London, featuring a live demo of web infrastructure enabling AI search, extraction, monitoring, and reasoning.
A 15-year quest to identify the obscure indie band Salvage, whose track 'Searchlights' charted once in 2006, finally ends after repeatedly throwing frontier AI models at the investigation — combining LLM reasoning, search tools, cover art analysis, and human follow-up across Discogs, Companies House, and GS1 databases.