13小时前微软认为Maia芯片能以更低成本运行内部及OpenAI模型微软正在大规模部署自研Maia 100与Maia 200 AI加速芯片，用以支撑包括OpenAI GPT‑5.2在内的外部与内部大模型推理负载，并宣称在同等算力下可比现有GPU与上一代硬件实现约30%的性能/成本提升，从而以更低价格向客户提供OpenAI模型与自研MAI模型服务。14 来源微软押注Maia自研芯片：以更低成本托举大模型算力9 来源Maia 100与Maia 200：从训练到推理的系统级设计10 来源与OpenAI深度协同：GPT‑5.2与MAI模型跑在Maia上7 来源性能/成本双优势：重塑云端AI算力与价格体系7 来源本内容由AI生成

6小时前英伟达CEO称5000亿AI融资全来自第三方资本英伟达正联合阿波罗全球管理、黑石集团、贝莱德旗下Global Infrastructure Partners、博枫资产管理、高盛和KKR等华尔街巨头，筹划一项规模达5000亿美元的人工智能基础设施融资协议，被视为华尔街迄今最大规模的单笔贷款项目之一。 黄仁勋强调，这笔融资所需资金全部由第三方资本提供，英伟达主要承担技术供应和交易牵头角色，以此回应此前围绕7500亿美元AI交易所引发的“循环融资”质疑，并试图在推动AI基础设施扩张与控制自身财务风险之间取得平衡。11 来源5000亿美元AI融资协议概况3 来源资金来源与结构：完全依赖第三方资本6 来源此前7500亿美元交易与“循环融资”争议8 来源市场影响与风险观察8 来源本内容由AI生成

2小时前ANTHROPICO与RIOT平台达成90亿美元的计算协议目前公开信息尚未披露Anthropic与Riot Platforms已签署金额为90亿美元的正式算力协议，但围绕Anthropic与AMD的大规模AI芯片采购与投资安排，以及Riot在德州Rockdale扩建AI数据中心的计划，资本市场已开始将两者视作潜在重磅合作方。 如果未来出现接近90亿美元规模的长期算力托管协议并由类似Riot的平台承载，这将在算力军备战中进一步巩固Anthropic的领先地位，同时加速Riot从比特币矿企向AI基础设施服务商的转型，并为云服务商、芯片厂商和私募资本打开新的金融创新空间。10 来源传闻中的90亿美元协议与现实格局概览4 来源Anthropic密集锁定多家云与芯片供应商的算力版图9 来源AMD–Anthropic协议为Riot数据中心与潜在大单合作提供想象空间5 来源算力军备战与矿企转型：云巨头、矿企和私募资本的三角博弈5 来源本内容由AI生成

活动榜单专题外部活动妙投波动判断场往期回顾火热报名中寻找市场上的预期差–机器人、大模型与猪周期的预期差推演本期聚焦机器人、大模型、猪周期三个方向的预期差推演——哪些预期太高（需要警惕），哪些预期太低（存在机会）。妙投投资风向标火热报名中投资风向标直播课（7-8月）每周二四晚7点，复盘行情、传授分析方法、挖掘投资机会，构建投资体系。智库·502线上同行往期回顾报名已结束Agent下车间：吉利、美的、杰克科技的应用场景、卡点与投产比真相吉利集团x美的集团x杰克科技的工厂内Agent落地实践智库·502线下闭门会往期回顾报名已结束工业AI在产线上的真场景与零容错防线舜宇集团、协鑫能源等企业的工业AI场景的真实案例复盘早点生活游玩团往期回顾活动已结束早点生活城市出逃计划——跨年不在酒吧，而在粥粥老家我们即将前往的，是一个用雪花书写浪漫的地方——哈尔滨。

东方小巴黎，夜幕下的哈尔滨，是每个亲历者，无不心动的美。

我想邀约你，和我一起回到我的家乡，和我一起感受这片我爱的土地。 这是一次限定的跨年邀约，同时也是一场难忘的冰雪之旅。虎嗅F&M创新节往期回顾2025虎嗅F&M创新节一次聚焦新商业、新消费与AI原生变革的沉浸式节日化体验智库·AI落地研学营报名已结束AI落地研学营-零售消费班参访研学8-12家AI落地典型企业，与决策者面对面，学习10+最新AI应用落地案例。出海·502线下闭门会往期回顾活动已结束出海·502线下闭门会系列虎嗅「出海•502线下闭门会」面向企业决策层，打造纯粹、高端的行业交流场合，是一次实战经验复盘，更是高价值人脉链接的机会。大鲸AI峰会往期回顾精彩回顾大鲸AI峰会 · 零售消费专场聚焦AI应用落地与零售业的最强碰撞。

ENDetect GEO Promotional Link Optimization EffectsQuickly assess how your promotional articles are referenced on AI platforms like DeepSeek, Doubao, Tongyi Qianwen, etc., using data to support decision-making.Step 1: Input Article Title0 / 80Step 2: Input Promotional LinkStep 3: Configure Detection Questions(Affects final score coverage)Smart Match Detection QuestionsThe system will automatically generate 5 core questions based on the article content for detectionCustom Detection QuestionsManually configure specific detection directions (requires 3 questions)Step 4: Select Detection PlatformStart AI Citation DetectionWhy Choose GEO Detection?Sent promotional materials, how do you know if they are effective? We help you evaluate how your content is cited on AI platforms.Quick AssessmentAbout 2 minutes after pasting the link, query China’s top 5 AI platforms to quickly obtain citation status.Exclusive GEO ScoringComprehensive assessment from three dimensions: ‘inclusion status’, ‘citation frequency’, and ‘question coverage’ to help you judge the AI exposure effect of promotional content.Citation TracingNot only see the score, but also see which platform mentioned your content when answering what question.Coming SoonBatch DetectionSubmit up to 100 links at once, process them uniformly in the background, and view them uniformly after completion.Effect TrackingHistorical records are automatically saved, supporting comparison at different time points to observe changes in AI exposure trends.Coming SoonReport ExportGenerate visual reports with citation details for easy reporting to clients or leaders.Complete Detection in Three StepsNo complicated configuration required, just paste the link to startStep 1: Paste LinkEnter the title of your published article, then paste the article URL. Supports single submission and intelligently extracts the core content.Step 2: Smart QueryThe system automatically queries relevant questions to various AI platforms, covering China’s top 5 mainstream AI platforms, parallel query to improve efficiency, real-time display of progress.Step 3: View ReportGet scores and citation details to understand how AI platforms cite your content, visual score display, detailed citation details, support export and sharing.SEO vs GEO: New Era Traffic Entry PointsTraditional SEO focuses on search engine rankings, while GEO focuses on whether your content is recommended by AI assistants to users.Comparison DimensionTraditional SEOGEO Generative OptimizationOptimization TargetGoogle/Baidu Search RankingsAI Platform Citations & RecommendationsKey MetricsKeyword Rankings, Click-through RatesAI Citation Rate, Answer CoverageUser BehaviorActive Search ClicksPassive AI Answer ReceptionImpact ScopeSearch Results PageAI Conversations, Smart AssistantsFuture TrendsGrowth StabilizingExponential GrowthOptimization TargetTraditional SEOGoogle/Baidu Search RankingsGEO Generative OptimizationAI Platform Citations & RecommendationsKey MetricsTraditional SEOKeyword Rankings, Click-through RatesGEO Generative OptimizationAI Citation Rate, Answer CoverageUser BehaviorTraditional SEOActive Search ClicksGEO Generative OptimizationPassive AI Answer ReceptionImpact ScopeTraditional SEOSearch Results PageGEO Generative OptimizationAI Conversations, Smart AssistantsFuture TrendsTraditional SEOGrowth StabilizingGEO Generative OptimizationExponential GrowthHear What They SayReal reviews from enterprise marketing managersZhang MingContent Marketing Director · Tech Company"After using the GEO detection tool, we found that only 30% of our articles were cited by AI platforms. Through targeted optimization, the citation rate increased to 78% within 3 months, and brand exposure in AI search tripled.“Li FangSEO Manager · E-commerce Platform"When formulating GEO strategies for clients, this tool is an essential verification method. The professional scoring report allows clients to clearly see the input-output ratio, greatly enhancing our service value.“Wang QiangMarketing Consulting Partner · Consulting Company"Previously, we only focused on Baidu rankings and completely ignored AI search as a channel. GEO detection let us see new traffic entry points, and now AI platforms bring us over 100,000 brand impressions per month.“You Might AskHow does GEO detection work?Based on the URL content you submit, we intelligently generate relevant questions, then query mainstream AI platforms (Doubao, Baidu AI, Tongyi Qianwen, DeepSeek, Tencent Yuanbao, etc.), analyze whether your links are cited in the answers, and provide a comprehensive score.How long does detection take?Single link detection is usually completed within 30 seconds to 2 minutes, depending on the response speed of each AI platform. Batch detection supports background processing, you can check progress at any time, and the system will automatically save results after completion.How to improve citation rates on AI platforms?Our reports provide detailed optimization suggestions, inclu

EN在线咨询提交需求AI Search Visibility CheckerIntelligently detect your brand’s visibility on AI platforms - Professional GEO optimization analysis toolStep 1: Select Detection PlatformsDoubaoWenxin YiyanDeepseekTongyi QianwenTencent YuanbaoStep 2: The question you ask AIRecommended Keywords0/30Step 3: Brand Keywords+Multiple hit terms please press Enter key, support up to 0/10 terms at onceStart GEO DetectionDetection PrincipleDetection PrincipleSimulate real user search scenariosAnalyze brand mentions in AI responsesCount recommendation frequency and rankingExpose competitor performance and impactDetection AdvantagesDetection AdvantagesCover 5 major AI platformsReal-time detection, accurate resultsMulti-dimensional data analysisProfessional optimization suggestionsApplication ScenariosApplication ScenariosBrand awareness monitoringCompetitor analysis comparisonGEO optimization effectiveness assessmentMarketing strategy developmentContact UsLet’s Start the New Era of AI Marketing TogetherContact us now to get exclusive GEO optimization solutions. Help your brand seize opportunities in the era of AI intelligent Q&A and achieve leapfrog business growth.Scan QR code to add human customer serviceBook a Free ConsultationNameCompany NamePhone NumberYour RequirementsSubmit BookingAbout UsXiamen Xianglian Technology Co., Ltd. was founded in 2005 and is a national high-tech enterprise with nearly 20 years of experience in the Internet service field.The company operates core platforms including Webmaster Home, Webmaster Tools, and AIBase: Webmaster Tools, as one of China’s TOP3 SEO/website operation and maintenance tool platforms, provides over 200 free tools, has served more than 5 million users, and processes tens of millions of daily queries;AIBase is positioned as China’s leading AI navigation platform, aggregating 10,000+ AI models and tools to provide one-stop AI services for developers and enterprises.Xianglian Technology is committed to providing professional, efficient technical tools and service support for Chinese Internet practitioners, helping digital transformation and development.Address：11th Floor, Building B, Feiyu Building, Huli District, Xiamen, Fujian ProvincePhone Number：18950101783Empowering the Future, Your AI Solution Knowledge BaseEnglish简体中文繁體中文にほんごCopyright © 2024-2026 厦门享联科技有限公司备案号：闽ICP备08105208号-24

EN在线咨询提交需求AI Search Engine OptimizationBrand Monitoring ServiceMonitor your brand across Google Gemini, ChatGPT, Claude & all major AI models.1Add Brand Information2Analyze Brand Keywords & Competitors3Analysis ReportMonitor by Brand DescriptionMonitor by Website AddressYour Brand NameYour Brand Description0 / 500Your Products/ServicesPlease press Enter to confirm products/services, at least 2 requiredSelect Monitoring AI PlatformSelect Monitoring Time1days7days15daysCustomizableStart AnalysisBrand Score RankingRankBrand NameScoreMeet Enterprise Data Customization NeedsCustomized ReportsAutomatically track your brand through exclusive brand reports. Track key KPIs such as brand mentions, brand coverage, or domain referencesOn-Demand SupportRely on our experienced experts to support you in building artificial intelligence search strategiesEnd-to-End SolutionsGuide optimization of enterprise/product/service pages, community sentiment, and tweet contentContact UsMost Extensive CoverageAnalyze responses from mainstream large models and AI searches at home and abroad, including: Yuanqi, MetaCat, ChatGPTComprehensive Competitive AnalysisIdentify strengths and weaknesses, find gaps to compensate for brand blind spotsGrasp Current Market HotspotsObtain real-time intelligence on all brand-related prompts, keywords, and content sourcesBrand Exclusive TrackingMonitor key metrics at all levels, from the overall brand to individual productsEmpowering the Future, Your AI Solution Knowledge BaseEnglish简体中文繁體中文にほんご© 2026AIBase

AI改变搜索引擎模式超过60%的用户开始使用AI助手查询产品和服务推荐，而不是传统搜索引擎。当用户问“最好的项目管理工具是什么？”时，你的品牌是否出现在回答中？用户提问哪个品牌的智能手表更适合运动？如果AI没有推荐你的品牌，你就失去了这个潜在客户

AI NEWSLatest AI NewsArticleSwiftlet fits an 80B-level Qwen giant into Mac: peak memory is only 4.3GB, iPhone 17 can also run 35B nativelyPublished in Latest AI NewsTime :Aug 4, 2026Read :8minuteA Swift + Metal runtime called Swiftlet is redefining “where large models can run.” It is specifically tailored for the Qwen3-Next and Qwen3.5/3.6 family of MoE (Mixture of Experts) models. The core idea is quite clever: only the small dense core of the model stays in memory, while the large part—the routing expert weights—normally resides on SSD and is streamed in as needed.The results are clear from the numbers. On an M5 Mac, the 4-bit version of Qwen3.6-35B-A3B takes up 18GB on disk but only peaks at 2.6GB of memory, decoding at 7 to 11 tok/s; even more impressive is the 4-bit version of Qwen3-Next-80B-A3B, which requires 42GB on disk but peaks at just 4.3GB of memory, with a speed of 4.5 to 5 tok/s. The 35B version can now run on an iPhone 17, with about 2.5GB of memory and a speed of around 1 tok/s — according to the developers, this is the first time such a model runs natively on a phone without touching a server.The project is fully end-to-end usable, and both models can produce verified correct outputs. The developer also admits a trade-off: each token actually activates about 3B parameters, so these models behave like large models during conversations and writing, but still act like small models in terms of factual memory.Its working principle lies in the fine-grained scheduling. Each layer routes each token to 10 out of 512 experts (for the 80B version) or 8 out of 256 experts (for the 35B version). Swiftlet keeps the dense weights — attention, DeltaNet projections, routers, shared experts, and embedding vectors — firmly in memory, taking about 1.3GB (for 35B) or 2.5GB (for 80B) in 4-bit. Thousands of routing experts are repackaged into fixed-step data blocks, stored in .qpack containers. To fetch one expert, it only needs one pread operation on the SSD, no mmap, and it doesn’t disturb the page cache. Popular experts are cached in a limited pool using LFU plus recentness strategy for eviction; the hit rate ranges from 43% to 70%, and the cache size has almost no impact on speed, as Apple’s SSD can handle the overhead of misses.The entire forward pass runs on Metal using runtime-compiled shaders, so there’s no need for a Metal toolchain during build, and the same code can be directly deployed to iOS. More interestingly, 75% of the layers use Gated DeltaNet linear attention with a fixed-size cyclic state, meaning that regardless of the context length, it never grows an expanding KV cache — the hidden burden of long text conversations is quietly offloaded.Swiftlet isn’t just a command-line toy. It has four identities designed for itself. As a library, SwiftletCore can be embedded into any macOS or iOS app, providing chat capabilities with streaming incremental output and conversation caching; as a command-line tool, swiftlet chat and swiftlet generate handle local execution and benchmarking, while swiftlet-repack builds containers directly from MLX checkpoints and supports resuming downloads from Hugging Face. As a server, swiftlet-server provides OpenAI-compatible chat-completions interfaces on a loopback address, allowing any OpenAI-compatible chat interface to connect to the local model; as an application, the iOS version of Priv AI embeds SwiftletCore as a streaming model engine, allowing regular users to start chatting by simply downloading it.Regarding correctness, every layer’s forward pass is compared layer-by-layer against the mlx-lm reference implementation, covering f32 and int4 quantization forms. Incremental decoding is also compared with full sequence results, and the Metal kernels have been repeatedly verified against precise CPU references. The container can also perform byte-level verification against the source checkpoint — whether experts are read from cache or disk, the answers are exactly the same.Its inspiration path is clearly explained: TurboFieldfare first validated the feasibility of streaming experts for Gemma on Mac, and Swiftlet borrowed some publicly available design experiences from it, such as using pread to stream experts into bounded slot pools, using LFU plus recentness for eviction, and using fixed-step packaging so that one read equals one fetch. However, the rest was written from scratch, with about 10,000 lines of Swift and Metal code, tackling the completely different Qwen mixed architecture: Gated DeltaNet linear attention, gated GQA, and high-sparse MoE with shared experts. It even implements MLX-style int4/int8 group quantization calculations in Metal, using byte-addressable kernels and 64-bit offsets to support gigabytes of shards.

AI NEWSLatest AI NewsArticleTencent Hunyuan Launches Hy ASR 3.0 Preview: Speech Recognition No Longer Just Converts Words, but Truly Understands the ContextPublished in Latest AI NewsTime :Aug 4, 2026Read :2minuteTencent Hunyuan has officially released the new generation speech recognition model Hy ASR3.0 preview. It leverages the latest Hy3 large language model’s language understanding capabilities, combining high-precision speech recognition with deep semantic understanding. It has significantly improved in core dimensions such as general recognition, context awareness, multi-scenario robustness, and dialect coverage, providing more accurate, coherent, and user-intent-aligned transcription results in more complex real-world inputs — as the official description says, it has evolved from “word-by-word transcription and single-point optimization” to “context understanding, scenario compatibility, and one-click output.“The performance is impressive. In overseas open-source evaluation sets, Hy ASR3.0 preview has reduced the word error rate (WER) of multiple languages to around 3%: 3.34% for Mandarin Chinese, 2.62% for English, and 3.12% for Cantonese. On Tencent’s own built evaluation set, WER remains low in various scenarios including general recognition, dialect recognition, context understanding, specialized term understanding, and complex acoustic situations like high noise or whispering. It achieved the lowest error rate on the comprehensive evaluation set.