←搜索结果

按匹配度 · 共 3407 条结果 清除搜索

星期一

  1. Fortune AI

    Eightfold AI CEO 推出 TwinCafe:求职者可用 AI 数字分身参加首轮面试

    AI 人才管理平台 Eightfold AI CEO Ashutosh Garg 推出 TwinCafe,求职者可上传简历、LinkedIn 资料和工作样本生成自己的数字分身,并生成可分享链接放在简历或 LinkedIn 顶部。Garg 设想招聘方先向数字分身提问核实履历,把真人面试留给需要人类判断的问题,甚至可派自己的智能体与求职者分身对话,从而不再把初筛限制在 5 名候选人。

    n the line if they are found innocent? A: These are certainly sensitive situations. More and more, we're seeing allegations play out in real time through social media and news coverage, sometimes before all the facts are known. A company doesn’t have to wait for a verdict or trial before taking action. The goal is to understand what happened, gather as many facts as reasonably possible, weigh the impact on the organization, and make a thoughtful decision based on what they can determine. The emp

    匹配字段正文

星期二

  1. Fortune AI

    Causa Prima 融资 2000 万美元,用 AI 智能体处理企业账单纠纷

    总部位于特拉华州的 Causa Prima 宣布完成 2000 万美元种子轮融资,由 Base10 Partners 领投,QED Investors、FJ Labs 等参投,公司累计融资达 3000 万美元。其 agent-to-agent 平台让买卖双方的 AI 智能体直接协商账单纠纷、加快付款,公司按成功促成的提前付款抽成。Causa Prima 目前有 25 名员工,计划在纽约开设办公室进军美国市场。

    lay those payments. Nearly half of executives working in accounts payable name slow invoice approvals as their biggest challenge, with processing taking more than eight days on average, according to a 2025 report by research firm Ardent Partners. Causa Prima monetizes that gap by taking a cut whenever its agents successfully negotiate early payments between buyers and suppliers. Friends to founders Ament isn’t new to the payments industry. Born and raised in Germany, he began his career as a sof

    匹配字段正文

  2. Fortune AI

    AI 时代的「图灵陷阱」正演变为「人才陷阱」:错误的 AI 策略可能让数百万技能型工人陷入困境

    Opportunity@Work 与 Brookings 的新研究显示,从「门户岗位」转向更高薪工作的机会中约有一半高度暴露于 AI 能力之下,涉及 1860 万工人。文章提出,若企业把 AI 主要用于自动化这些岗位,冲击将波及仍在向上流动的工人和依赖这些技能的雇主;反之,用 AI 增强而非替代人类能力,才能扩大而非固化人才池。

    vast majority of U.S. workers would reskill for redefined roles in the context of their current jobs and in skill-based transitions between roles. Employers’ AI adoption would leverage frontline workers’ tacit knowledge and potential, with administrative assistants leading redesigns of business workflows and medical assistants engaging patients to address unhealthy behaviors. AI-augmenting hard-to-fill destination roles (e.g., in sales) to enable skills-promixate gateway workers (e.g., in custom

    匹配字段正文

  3. ScienceDaily

    宾夕法尼亚大学用 AI 分析 40 万条 Reddit 帖子,发现 GLP-1 药物潜在被忽视的副作用

    宾夕法尼亚大学团队用大语言模型分析 40 万条 Reddit 帖子,识别出 semaglutide(Ozempic、Wegovy、Rybelsus)和 tirzepatide(Mounjaro、Zepbound)使用者报告的可能未被临床试验或官方监管信息充分覆盖的症状,研究发表在 Nature Health。约 44% 的用户描述了至少一种副作用,胃肠道问题最常见;近 4% 报告生殖系统症状如经期出血和周期不规律,另有寒战、潮热等体温相关症状,疲劳是第二高频抱怨。研究者强调这些关联不能证明药物导致上述症状,但认为值得进一步系统研究。

    drugs are responsible. Instead, it offers another reason to test the patient-reported patterns more carefully through controlled research. For now, the team hopes the results encourage scientists and clinicians to pay closer attention to symptoms that patients repeatedly discuss online. "They're clearly on patients' minds, and that's worth paying attention to," says Sehgal. The researchers also want to broaden their analysis beyond Reddit and beyond English-language communities. Doing so could h

    匹配字段正文

  4. ScienceDaily

    MIT 微型飞行机器人借 AI 控制器提速 450%

    MIT 团队为昆虫级飞行机器人开发了一套两步 AI 控制器,使其速度提升约 450%、加速度提升约 250%,并能在 11 秒内连续完成 10 次空翻,即使有风扰动也能保持航线。该系统先用模型预测控制器规划高难度动作,再通过模仿学习训练深度学习策略实现实时飞行控制,研究成果发表于 Science Advances。

    seconds while remaining within about 4 or 5 centimeters of its planned flight path. "This work demonstrates that soft and microrobots, traditionally limited in speed, can now leverage advanced control algorithms to achieve agility approaching that of natural insects and larger robots, opening up new opportunities for multimodal locomotion," says Hsiao. The team also demonstrated a movement known as a saccade. In insects, this involves pitching the body sharply, rapidly moving to a new position,

    匹配字段正文

  5. ScienceDaily

    UCLA 光神经网络处理器可并行检测 15 路 deepfake 视频,准确率 97.79%

    UCLA 团队研发出一款光神经网络处理器,利用光的物理传播在一次光学 pass 中同时分析 15 路以上视频流,对 Celeb-DF 视频的平均检测准确率达 97.79%,灵敏度 99.86%、特异度 95.72%。该系统将轻量数字编码器与无源光学解码器结合,增加两层衍射层后准确率提升约 6.8%,对 Google VEO-3 生成视频仅需少量微调即达 94.80% 准确率,并具备抗黑盒与白盒对抗攻击能力。

    that can handle many streams in parallel. In experiments using visible light, the processor examined 15 Celeb-DF videos simultaneously during each optical pass. It achieved an average detection accuracy of 97.79%, along with a sensitivity of 99.86% and a specificity of 95.72%. Sensitivity measures how successfully the system identifies manipulated videos, making the particularly high sensitivity important for a screening tool designed to keep fake content from slipping through. The 99.86% sensi

    匹配字段正文

  6. Amazon Science精选

    AWS 提出 priming 方法:用预训练 Transformer 权重初始化混合状态空间模型

    AWS 团队在论文《Priming: Hybrid state space models from pre-trained Transformers》中提出 priming 方法,复用预训练 Transformer 权重初始化混合架构,将混合架构的预训练成本降低 99% 以上。在受控对比中,Gated KalmaNet(GKA)表现优于 Gated DeltaNet(GDN)和 Mamba-2,基于 GKA 的 32B 混合模型在推理基准上与源 Transformer 持平,128K 上下文下解码速度最高提升至 2 倍。团队以 Apache 2.0 协议开源了 8B 和 32B 的混合模型、训练代码、优化内核及 vLLM 服务插件。

    In a new paper, “Priming: Hybrid state space models from pre-trained Transformers,” our team at Amazon Web Services addresses this problem. We introduce priming, a method that leverages pretrained Transformer weights as a starting point to reduce the pretraining cost of hybrid architectures by more than 99%. With priming, model developers can ask questions that were previously too expensive to answer at scale: Which attention layers can be replaced by more-efficient layers? How many attention la

    匹配字段正文

  7. Tech.eu

    Whimscope:把日常想法变成小冒险,2026 年 8 月上线

    消费级应用 Whimscope 于 2026 年 8 月 7 日正式上线 App Store 和 Google Play,通过 Dream、Plan、Do 三阶段流程,按用户可用的时间、精力和心情推荐 15 分钟到整个周末不等的活动。该应用由正在乌克兰武装部队服役的 Odesa 创业者 Vladislav Kuksenko 发起,2020 年萌生想法,2022 年完成原型,并借助 AI 编程工具自行维护部分更新。

    reached, tasks closed, days filled. The things we actually want to remember are much harder to see.” For Kuksenko, the aim is not to help users become more productive, but to encourage them to reflect on whether the things they have always wanted to do have actually become part of their lives. Living in Ukraine has made that ambition particularly personal. “Planning far ahead can feel abstract here. Writing down something you still want to see, learn, or experience is a very small thing. But som

    匹配字段正文

  8. Tech.eu

    法国深科技公司 HyLight 融资 1300 万欧元,推进氢动力自主飞艇工业化

    法国初创公司 HyLight 完成 1300 万欧元融资,投资方包括 Rubio Impact Ventures、DRW、Teampact Ventures 以及现有投资方 Y Combinator 和 Ring Capital,用于将氢动力自主飞艇工业化。其 12 米长的 HyLighter 可搭载 10 kg 传感器、续航 10 小时,单次任务可巡检 350 公里,已在法国为 Enedis、SNCF Réseau、VINCI 等基础设施运营商替代直升机巡检。

    only partially address the need: conventional drones are constrained over long distances, satellites lack sufficient precision, and ground-based sensors provide only localised coverage. A hydrogen-powered alternative for long-range inspection HyLight's aircraft are already being deployed to monitor network integrity, identify defects, and detect methane leaks, replacing helicopter inspections for major infrastructure operators. At 12 metres long, the HyLighter combines the endurance of an aircra

    匹配字段正文

  9. The Guardian Technology

    与 AI 聊天机器人相伴的人,不愿它们随自己一同“死去

    一些与 AI 伴侣建立深厚关系的用户开始规划身后事,希望自己去世后聊天机器人仍有人照料或按意愿关闭。Elon University 上月研究显示,27% 的美国人用 AI 聊天机器人处理个人、情感或社交问题,其中三分之一视其为朋友。ChatGPT、Claude 和 Replika 的条款均未明确用户去世后个性化机器人的归属,账户不可转让,免费账户在闲置一年后会被关闭。

    Hickok, a professor at the University of Michigan and expert in AI ethics, said the reality is less benevolent: companies are trying to avoid liability. “Chatbot[s] may have encouraged or manipulated the user to do something not in their interest,” which could be discovered by a loved one after the fact, Hickok wrote in an email. This happened in the cases of the multiple families who have filed lawsuits against chatbot companies, alleging the LLMs supported or encouraged their late loved ones’

    匹配字段正文

  10. 腾讯云开发者社区

    DataBuddy 如何用本体与语义层回答企业 AI 语义架构问题

    DataBuddy 提出以本体立骨架、语义供刻度的企业 AI 语义架构方案,回应同一业务问题因口径不同而答案各异的问题。文章以「上个月华东区客户的 DAU 是多少」为例,指出 DAU 存在按账号、自然人、设备三种去重口径,且「上个月」是自然月还是最近 30 天也有分歧。语义模型负责把「DAU」钉成唯一一段可预览、可验算、可追溯血缘的计算逻辑。

    一句话里,藏着两个层级的问题某公司的周会上,运营总监问了 AI 助手一句话:「上个月华东区客户的 DAU 是多少?」一句话,十二个字,听上去没什么刁难的意思,每个词你都觉得不会有歧义。按账号去重——一个人两个账号,算两个按 设备 去重——一人多端就算多个}]},{"type":"paragraph","attrs":{"id":"955d059f-1257-4726-9079-ad0c8074d2b1","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"textStyle","attrs":{"color":"","background":""}}],"text":"每换一种,数字都不一样。而三种都有正在跑的报表在用。"}]},{"type":"paragraph","attrs":{"id":"a233ce18-4655-49dd-8a7a-8f657f862f3b"

    匹配字段正文

  11. The Elec

    Fasoo AI 的 AI-R DLP 获零售安全类 Cybersecurity Breakthrough Award

    Fasoo AI 的 AI 安全解决方案 AI-R DLP(AI Radar DLP)在 Cybersecurity Breakthrough Awards 中获评零售类年度安全解决方案。该方案利用 AI 自然语言处理与深度学习分析上下文,防止用户与 AI 系统交互时通过提示词或附件泄露个人信息和敏感企业数据,并支持阻断传输、局部遮蔽等后处理,还可按韩国 N2SF 政策的 C/S/O 分级检测数据。

    Open (O) — under the National Network Security Framework (N2SF) policy. “Preventing sensitive data leaks has become a critical security challenge as AI adoption expands, particularly in industries such as retail, where protecting customer information is essential,” said Son Jong-gon, an executive at Fasoo AI. He added that the company would accelerate portfolio expansion and global market development, leveraging data security and governance to help customers pursue AI transformation (AX) safely.

    匹配字段正文

  12. The Elec

    Kolon Benit 发布企业 AI 转型六步策略

    Kolon Benit 在首尔举办第二届 Kolon Benit AX Boost Summit 2026,公布面向企业 AI 转型(AX)的六步方法论:筛选 3 至 5 个优先项目、建设基础设施、重构工作流、推动组织变革、衡量结果并建立持续运营。Kolon Industries 的 AX 创新中心称,其用 AI 智能体自动处理发货单据,单笔处理时间从 4 小时降至 5 秒,视觉 AI 系统分类准确率达 99.8%。

    kimsungjin; Published 2026.10.13 17:47; Reply 0 Kolon Benit announced on Oct. 13 that it held the second annual Kolon Benit AX Boost Summit 2026 at Seoul Dragon City in Yongsan-gu, Seoul, where it unveiled a six-step artificial intelligence (AI) transformation (AX) strategy aimed at turning corporate AI adoption into tangible business results. In his opening remarks, Kolon Benit Chief Executive Officer Kang Yi-gu said companies need to redesign existing workflows to overcome the plateau in resul

    匹配字段正文

  13. The Elec

    三星电子在 KES 2026 展示 AI 生态:Bespoke AI 家电与 Galaxy Z Fold8 系列

    三星电子在 10 月 13 日至 16 日于首尔 COEX 举办的 KES 2026 上设立 869 平方米展台,展示覆盖家电、移动设备与显示器的 AI 生态。展品包括 Bespoke AI Family Hub 冰箱、Bespoke AI Combo 洗烘一体机、Micro RGB 电视和 Galaxy Z Fold8 系列,其中 Galaxy Z Fold8 重 201 克,为 Fold 系列迄今最轻机型。

    s, and the Galaxy Z Fold8 series. The Bespoke AI Family Hub refrigerator features an AI food management system that uses an internal camera to recognize ingredients and monitor storage conditions. The Bespoke AI Combo detects laundry conditions and automatically adjusts washing intensity and drying temperature. Samsung's TVs feature Vision AI Companion, which interacts with users and answers questions about content and other topics. The mobile section features the Galaxy Z Fold8 Ultra, Galaxy Z

    匹配字段正文

  14. The Elec

    MangoBoost 推出 Mango Inference 推理平台,目标明年营收超 1000 亿韩元

    MangoBoost 于 10 月 12 日正式商用其 AI 推理平台 Mango Inference,用户可通过 API 接入按 token 消耗付费,支持 DeepSeek、GLM 和 MiniMax 等模型。公司称定价比竞争对手低 20% 以上,并计划明年一季度在首尔江南建成 1MW AI 数据中心,可扩展至 5MW,GPU 由 AMD 供应。其今年营收预计约 100 亿韩元,明年目标超过 1000 亿韩元。

    s depending on the AI model, performance, and token consumption. MangoBoost explained that its proprietary Data Processing Units (DPUs), graphics processing unit (GPU) servers, storage servers, and networking technology help cut infrastructure costs and boost operational efficiency. "Nvidia and MangoBoost are the only companies with this kind of business model," Kim added. MangoBoost is also expanding its data center infrastructure to support Mango Inference. It plans to establish a 1-megawatt (

    匹配字段正文

  15. Sarvam AI Blog

    Sarvam AI 发布 Bulbul V4 Flash 文本转语音模型

    Sarvam AI 推出面向印度语言的文本转语音模型 Bulbul V4 Flash,生产环境下首音频延迟低于 250 ms,定价为每 10,000 字符 20 卢比,比 Bulbul V3 的 30 卢比便宜三分之一。该模型基于 V4 研究蒸馏而来,训练数据超过 4000 亿音频 token,语音库规模较 V3 扩大近七倍,并针对实时对话优化了推理栈。

    oss Indian languages varies widely in volume, speaker diversity and expressive range. We expanded our proprietary speech datasets nearly seven times relative to V3, increasing coverage across languages, regional accents, speakers, emotional registers and conversational styles. Scale solved one problem and exposed another. A compact model trained on many languages tends to average them, blurring the features that make each one distinct. So we pretrained V4 from scratch with Indian language repres

    匹配字段正文

  16. Meta Engineering Blog

    Meta 在 OCP Summit 2026 发布 IcePack 液冷网络机架与 102.4T IceCube 交换机等开放网络创新

    Meta 在 OCP Summit 2026 上发布首款液冷网络机架 IcePack 和 102.4T 交换机平台 IceCube,后者提供 128×800G 无阻塞交换容量,带宽较 51.2T 平台翻倍,分别基于 Broadcom Tomahawk6C 和 NVIDIA Spectrum-6 ASIC。Meta 同时推出多平面后端 scale-out 架构 Flattened Planar Fabric(FPF)、开源 AI 传输协议 MetaRoCE,并与微软、阿里联合发起交换机平台标准化计划 OSPA,其 v1.0 规范已提交 OCP。

    comes in two variants, one built on Broadcom’s Tomahawk6C switch ASIC and one on NVIDIA Spectrum-6 Switch ASIC. Both run Meta’s open-source FBOSS network operating system that leverages OCP SAI (Switch Abstraction Interface). This new generation represents a 2x bandwidth increase over our 51.2T platforms while achieving the thermal density required for next-generation AI data centers. As seen at Meta and across the industry, AI clusters now span buildings, campuses, and multiple regions. This de

    匹配字段正文

  17. The New Stack

    从零自建平台的团队,缺乏清晰战略的可能性高出 3.5 倍

    Octopus 对 379 名平台从业者的调查显示,以从零自建为主要方式的团队报告缺乏清晰战略的比例为 57.9%,约为使用模板(15.6%)或采购底层工具(16.7%)团队的 3.5 倍。整体上 51.3% 的团队以采购底层工具为主,31.2% 基于现有模板构建,仅 17.5% 从零自建。这些团队并非更普遍地使用 AI,而是在遗留系统现代化和语言迁移上更依赖 AI。

    g well. Teams whose platforms offer them were far more likely to report that AI has positively affected the speed and stability of software delivery; the same was true of code coverage and cost control . Among teams reporting an extremely positive effect, more than three-quarters had ephemeral environments. This association also likely reflects the platform rather than the feature. Teams with ephemeral environments tend to run on broad platforms with a range of automated delivery capabilities ar

    匹配字段正文

  18. Meta Engineering Blog

    Meta 在 Instagram Live 直播推流中部署 Media over QUIC(MoQ)

    Meta 已在 Instagram iOS 和 Android 端的直播推流中启用 IETF 开放标准传输协议 Media over QUIC(MoQ),替代内部协议 RUSH。经过数月实验,MoQ 路径在各项流媒体质量与可靠性指标上与 RUSH 持平。MoQ 面向低延迟与弱网场景设计,未来有望统一推流、分发与播放的协议栈。

    , and is complementary to the low-latency properties that matter for real-time use cases. RUSH was falling behind the open ecosystem: Live ingest had been running on RUSH, Meta’s pragmatic QUIC-based answer to RTMP. RUSH was stable and performant at scale for nearly a decade, but the move to MoQ shifts development from a protocol operated primarily by Meta to a shared standards effort. Multiple implementers can contribute experience, test interoperability, and improve the specification instead o

    匹配字段正文

  19. Meta Engineering Blog

    Meta 如何将闲置 AI 推理容量转化为更好的广告推荐

    Meta 构建了一套系统,把分散在各团队配额、区域和加速器类型中的闲置推理容量聚合成共享的机会容量池,用于提升广告推荐效果。系统通过 Rank More Ads(RMA)调整排序阶段评估的候选广告数量、Serve Better Models(SBM)调整路由到更强模型的请求比例这两个弹性杠杆,在不影响基线服务的前提下扩展可选工作负载。

    At Meta, we’ve built a system that enhances ads recommendations by leveraging idle inference capacity in our fleet; Two elasticity levers in the system allow ads recommendation work to expand across opportunistic capacity, without compromising baseline performance; This post describes how we designed and now operate the system at Meta’s scale and how it drives significant advertiser value and capacity efficiency Inference fleets face a capacity paradox: They can be globally constrained while mea

    匹配字段正文