←搜索结果

按匹配度 · 共 3407 条结果 清除搜索

星期六

  1. TechCrunch AI

    盘点能住进你短信里的 AI 智能体:Caddy、Comma、Fambot、Folk、Instinct 等

    越来越多 AI 智能体无需下载 App,直接通过 iMessage、RCS、SMS、WhatsApp 或 Telegram 收发消息,即可记住上下文、连接已有应用并代为完成任务。Instinct 在 2026 年 9 月以 10 亿美元融资、100 亿美元估值成为焦点,此前曾以 25 亿美元估值融资 3.5 亿美元;Folk 提供 8.33 美元/月的 Pro 订阅,Martin 订阅 21 美元/月起,Comma 免费且开源,Fambot 已融资 350 万美元 pre-seed。这些智能体覆盖日程、邮件、旅行预订、家庭事务等场景。

    th service providers, and help coordinate responsibilities beyond the immediate household. Orbits is backed by investors like Andreessen Horowitz (a16z) Speedrun fund, N49P, and Garage Capital. Pally Users can interact with Pally as they would with a friend while giving it access to the apps they already rely on. The assistant can connect information spread across services such as WhatsApp, Gmail, calendars, and Google Drive, helping users keep track of plans, receipts, dates, files, and other d

    匹配字段正文

  2. The Verge AI

    AI 智能体厂商竞相承诺隐私保护,能否兑现存疑

    The Verge 报道称,OpenAI 在 DevDay 发布 AI 智能体 Dots 并承诺为前沿 AI 隐私设定新标准,同时暗讽 Meta 的 Muse 未能保护用户数据。但 Muse 此前也以隐私安全为卖点,上线后却出现零日漏洞、默认用用户数据训练模型等问题,Meta 自身仍可访问用户数据。Dots 目前仅面向每月 100 美元及以上的 ChatGPT 订阅档位开放,隐私争议较少,但用户仍可能不愿向 AI 智能体交出大量个人信息。

    and easy to use and scale to billions of people.” User data is stored on a secure VM, what Zuckerberg described as an “isolated linux computer with a browser, CPU, memory, and storage.” The company said that most of the effort in building Muse was related to “careful design and engineering to operate [it] more safely.” Meta’s blog post continued, “Muse can and will still make mistakes, but we expect they’ll be much less frequent and cause much less damage due to the safety systems we’ve built i

    匹配字段正文

  3. The New Stack

    微软在 Foundry 发布基于阿里 Qwen 的决策模型 Microsoft-Decision-1

    微软在 Foundry 上线 Microsoft-Decision-1,该决策模型基于阿里 Qwen3.5-9B 后训练而成,而非采用合作伙伴 OpenAI 的模型,定价为每百万输入 token 0.042 美元、输出免费。微软称其已在内部四个团队测试,用于整理超 1 万条 Xbox 玩家反馈、为 Copilot 打分、事故现场拉取上下文等场景,并称在 Xbox 场景下比 GPT-6 Sol 快 14 倍以上。该模型仅支持文本、上限 32,768 token 并返回 JSON,不支持图像,微软尚未公布开放权重。

    testing approach. Microsoft tested Decision-1 with eight kinds of perturbations, including reordered options and paraphrased descriptions, which changed 1.3% of its answers on average. JevOut didn’t test Decision-1, so it’s unclear how the model would hold up against similar attacks. Microsoft also hasn’t confirmed whether Decision-1 fully supports the System One API competitors adopted, leaving questions about interoperability and how much trust to place in its confidence scores. The post Micr

    匹配字段正文

  4. Robotics & Automation News

    纽约州延长与 Intuition Robotics 的 ElliQ AI 陪伴计划

    纽约州老龄办公室(NYSOFA)延长与 Intuition Robotics 的合作,继续为有孤独和社会隔离风险的纽约老年人提供 ElliQ 主动式 AI 陪伴设备。2026 年项目数据显示,94% 的用户表示孤独感减轻,97% 表示整体健康状况改善,用户日均与 ElliQ 互动 48 次,高于此前报告的 29 次。该项目自 2022 年启动,2025 年有超过 3500 名老年人申请参与。

    am data from 2026 demonstrates ElliQ’s impact on the older adults living with the device: 92 percent of ElliQ users find the companion useful; Participants engage with ElliQ an average of 48 times per day, up from 29 daily interactions in reporting from previous years; 94 percent report feeling less lonely, 97 percent report improved overall wellbeing, 88 percent say ElliQ has made a positive difference in their daily lives, and 79 percent feel more connected to the world around them; Every mont

    匹配字段正文

  5. Apple Machine Learning Research

    Apple 机器学习研究 2026 年论文列表:NeurIPS、EMNLP 等会议成果

    Apple 机器学习研究页面列出 2026 年论文,涵盖 NeurIPS、EMNLP 等会议,主题包括 Normalizing Trajectory Models、RISED 智能体多环境选择与自蒸馏、扩散模型置信度极限、强智能体自主 ML 工程所需 harness 程度,以及 LLM 条件化的有效性-流畅性权衡。

    oders via Latent-Space Distillationresearch area Methods and Algorithms, research area Speech and Natural Language ProcessingPublished year 2026AuthorsPrasanth Yadla‡, Mohammad Samragh Razlighi‡, Dongseong Hwang, Mingbin Xu, Yuanyuan Zhang, Chung-Cheng Chiu, Yongqiang Wang† , Yuan Liu§ , Zhen Huang, Xiaodan Zhuang; How to Guide Your Language Flowresearch area Methods and AlgorithmsPublished year 2026AuthorsRohit Dilip† , Tianrong Chen, Yuyang Wang, David Van Valen†, Josh Susskind, Miguel Angel B

    匹配字段正文

  6. Apple Machine Learning Research

    Apple Machine Learning Research 公布 2026 年 NeurIPS 等会议论文列表

    Apple Machine Learning Research 页面列出 2026 年多篇论文,包括入选 NeurIPS 的 Normalizing Trajectory Models、RLTL;DR: Self-Improvement by Internalizing Self-Generated Feedback 和 The Communication Bottleneck: A Round-Trip Study of Tree-Structured Expression Serialization in Language Models,以及入选 EMNLP 的 On the Effectiveness-Fluency Trade-Off in LLM Conditioning。其余论文涵盖扩散模型置信度、智能体自主 ML 工程、多语言语音模型等方向。

    oders via Latent-Space Distillationresearch area Methods and Algorithms, research area Speech and Natural Language ProcessingPublished year 2026AuthorsPrasanth Yadla‡, Mohammad Samragh Razlighi‡, Dongseong Hwang, Mingbin Xu, Yuanyuan Zhang, Chung-Cheng Chiu, Yongqiang Wang† , Yuan Liu§ , Zhen Huang, Xiaodan Zhuang; How to Guide Your Language Flowresearch area Methods and AlgorithmsPublished year 2026AuthorsRohit Dilip† , Tianrong Chen, Yuyang Wang, David Van Valen†, Josh Susskind, Miguel Angel B

    匹配字段正文

  7. Apple Machine Learning Research

    Apple Machine Learning Research 发布 2026 年论文列表

    Apple Machine Learning Research 公布 2026 年论文列表,涵盖 NeurIPS、EMNLP 等会议,涉及计算机视觉、语音与自然语言处理、方法算法等方向。论文包括 Normalizing Trajectory Models、RISED、SCLATE 等,作者来自 Apple 及合作机构。

    oders via Latent-Space Distillationresearch area Methods and Algorithms, research area Speech and Natural Language ProcessingPublished year 2026AuthorsPrasanth Yadla‡, Mohammad Samragh Razlighi‡, Dongseong Hwang, Mingbin Xu, Yuanyuan Zhang, Chung-Cheng Chiu, Yongqiang Wang† , Yuan Liu§ , Zhen Huang, Xiaodan Zhuang; How to Guide Your Language Flowresearch area Methods and AlgorithmsPublished year 2026AuthorsRohit Dilip† , Tianrong Chen, Yuyang Wang, David Van Valen†, Josh Susskind, Miguel Angel B

    匹配字段正文

  8. Apple Machine Learning Research

    Apple 机器学习研究发布 2026 年 NeurIPS、EMNLP 等论文列表

    Apple 机器学习研究页面列出 2026 年多篇论文,包括发表于 NeurIPS 的 Normalizing Trajectory Models、RLTL;DR 和 The Communication Bottleneck,以及发表于 EMNLP 的 On the Effectiveness-Fluency Trade-Off in LLM Conditioning。其余论文涵盖 RISED 智能体多环境选择与自蒸馏、扩散模型置信度极限、多语言语音模型语言判别、SCLATE 持续学习智能体训练与评估等方向。

    oders via Latent-Space Distillationresearch area Methods and Algorithms, research area Speech and Natural Language ProcessingPublished year 2026AuthorsPrasanth Yadla‡, Mohammad Samragh Razlighi‡, Dongseong Hwang, Mingbin Xu, Yuanyuan Zhang, Chung-Cheng Chiu, Yongqiang Wang† , Yuan Liu§ , Zhen Huang, Xiaodan Zhuang; How to Guide Your Language Flowresearch area Methods and AlgorithmsPublished year 2026AuthorsRohit Dilip† , Tianrong Chen, Yuyang Wang, David Van Valen†, Josh Susskind, Miguel Angel B

    匹配字段正文

  9. Apple Machine Learning Research

    Apple Machine Learning Research 发布 2026 年论文列表

    Apple Machine Learning Research 公布 2026 年论文列表,涵盖 NeurIPS、EMNLP 等会议,主题包括 Normalizing Trajectory Models、RISED 智能体多环境选择与自蒸馏、扩散模型置信度极限,以及多语言语音模型的语言判别等。

    oders via Latent-Space Distillationresearch area Methods and Algorithms, research area Speech and Natural Language ProcessingPublished year 2026AuthorsPrasanth Yadla‡, Mohammad Samragh Razlighi‡, Dongseong Hwang, Mingbin Xu, Yuanyuan Zhang, Chung-Cheng Chiu, Yongqiang Wang† , Yuan Liu§ , Zhen Huang, Xiaodan Zhuang; How to Guide Your Language Flowresearch area Methods and AlgorithmsPublished year 2026AuthorsRohit Dilip† , Tianrong Chen, Yuyang Wang, David Van Valen†, Josh Susskind, Miguel Angel B

    匹配字段正文

  10. Apple Machine Learning Research

    Apple Machine Learning Research 发布 2026 年论文列表

    Apple Machine Learning Research 公布 2026 年论文列表,涵盖 NeurIPS、EMNLP 等会议,主题包括 Computer Vision、Methods and Algorithms、Speech and Natural Language Processing 等。其中 Normalizing Trajectory Models 与 RLTL;DR 入选 NeurIPS,RISED 聚焦智能体多环境选择与自蒸馏。

    oders via Latent-Space Distillationresearch area Methods and Algorithms, research area Speech and Natural Language ProcessingPublished year 2026AuthorsPrasanth Yadla‡, Mohammad Samragh Razlighi‡, Dongseong Hwang, Mingbin Xu, Yuanyuan Zhang, Chung-Cheng Chiu, Yongqiang Wang† , Yuan Liu§ , Zhen Huang, Xiaodan Zhuang; How to Guide Your Language Flowresearch area Methods and AlgorithmsPublished year 2026AuthorsRohit Dilip† , Tianrong Chen, Yuyang Wang, David Van Valen†, Josh Susskind, Miguel Angel B

    匹配字段正文

  11. Apple Machine Learning Research

    Apple 机器学习研究发布 2026 年论文列表

    Apple 机器学习研究(Apple Machine Learning Research)官网公开了 2026 年论文列表,涵盖 NeurIPS、EMNLP 等会议,涉及计算机视觉、方法算法、语音与自然语言处理、人机交互等方向。其中《Normalizing Trajectory Models》与《RLTL;DR: Self-Improvement by Internalizing Self-Generated Feedback》均发表于 NeurIPS。

    oders via Latent-Space Distillationresearch area Methods and Algorithms, research area Speech and Natural Language ProcessingPublished year 2026AuthorsPrasanth Yadla‡, Mohammad Samragh Razlighi‡, Dongseong Hwang, Mingbin Xu, Yuanyuan Zhang, Chung-Cheng Chiu, Yongqiang Wang† , Yuan Liu§ , Zhen Huang, Xiaodan Zhuang; How to Guide Your Language Flowresearch area Methods and AlgorithmsPublished year 2026AuthorsRohit Dilip† , Tianrong Chen, Yuyang Wang, David Van Valen†, Josh Susskind, Miguel Angel B

    匹配字段正文

  12. X.PIN精选

    中国将出台国家级计划应对 AI 就业与技能挑战

    国务院新闻办公室表示,中国将出台国家级计划,以促进就业并提升劳动力技能,应对 AI 发展。该计划旨在通过 AI 驱动的创新创造就业、鼓励传统行业采用 AI,并帮助劳动者建立 AI 技能,将纳入下一个五年规划。文中还提到,与美国不同,中国似乎更侧重实际应用,而非对 AI 导致失业的担忧。

    China will roll out a national plan to boost employment and workforce skills in response to AI development, according to the State Council Information Office. The plan aims to create jobs through AI-driven innovation, encourage AI adoption in traditional industries, and help workers build AI skills. This will be part of China's next five-year-plan. Unlike the US, China seems to be focusing heavily on practical applications rather than fear about AI job losses.

    匹配字段正文

  13. The Decoder

    a16z 第七版消费级 AI 榜单:首次追踪美国消费者实际支出

    a16z 发布第七版最常用消费级 AI 产品榜单,首次借助 YipitData 数据追踪美国消费者信用卡上的实际 AI 支出。数据显示近半数美国消费者使用 AI,但仅 4.5% 在 8 月拥有 ChatGPT、Gemini 或 Claude 的付费个人订阅;支出高度集中,前 1% 用户贡献近五分之一支出,月均约 900 美元,而典型付费用户约 25 美元。a16z 指出,付费者更偏好 n8n、Manus、fal 等构建与自动化工具,专业用途是消费级 AI 的首个真实付费市场。

    extremely concentrated. The top one percent of spenders accounts for nearly a fifth of all observed AI spending, more than the entire bottom half combined. That top one percent averages about $900 per month and keeps climbing, while the typical paying user spends about $25 and has held steady at that level for a while. Their product choices reveal who they are. They favor building and automation tools like n8n, Manus , or fal at much higher rates, along with creative tools like Higgsfield, Figma

    匹配字段正文

  14. Latent Space

    TypeSafe/Jev 上线三周 ARR 破 1 亿美元、估值 75 亿美元

    TypeSafe 的 Jev 上线三周即实现超 1 亿美元 ARR、估值达 75 亿美元,Sequoia 称其首周就跨过 1 亿美元 ARR。同日多家厂商推出返回概率、选项或分数的“决策”模型:OpenAI Decisions API 基于 GPT-6 Luna、输入 $0.10/M tokens 且无输出费用,Perplexity pplx-decider-v1.1-27b 在 1,071 个案例上取得 94.5% 的 Decision Bench 准确率。Unsloth 发布免费 notebook,可在 8GB 显存上把 Qwen3.5-4B 训练成决策模型。

    ward pass that shares the KV cache; Results: Per-turn reasoning latency falls by more than 90%, but end-to-end latency per question falls only 28–54%. On Qwen3-VL-4B with GRPO, average success is 61.37% versus 61.25% for a TAPO+GSPO baseline Claude Managed Agents dynamic workflows (public beta): A lead agent writes a phased plan, fans it out to up to 1,000 agents per run, then merges the results. It is enabled with multiagent 20261001 (@ClaudeDevs, config).Cost warning: Anthropic advises startin

    匹配字段正文

  15. MarkTechPost

    微软发布 Microsoft-Decision-1:基于 Qwen3.5-9B 的决策打分模型

    微软发布 Microsoft-Decision-1,一个基于阿里 Qwen3.5-9B 后训练的决策打分模型,对固定选项返回校准后的概率而非生成文本,现已上线 Microsoft Foundry 和 OpenRouter。它支持 32,768 token 上下文,在微软 36 项基准对比中平均准确率 83.5%,p50 延迟 85 ms,输入定价为每百万 token 0.042 美元、输出免费。模型仅提供托管 API,不开放权重,且所有基准均为厂商自测。

    8-token context window; Runs on: Hosted API only (Microsoft Foundry, OpenRouter via Azure). No open weights, no quantized variants, hardware not disclosed; Performance: Highest average accuracy in Microsoft’s 36-benchmark comparison, at 85 ms p50 latency; Best: 83.5% average accuracy across 36 benchmarks, ahead of Quyet-1.0-Large at 81.9%; Worst: Calibration of 92.2, second to Quyet-1.0-Large at 93.1. Text-only, no explanations; Bottom line: Best thing: fast, cheap, calibrated decisions at $0.04

    匹配字段正文

  16. Descript 官方产品更新

    Descript 指南:如何从零开始创建 YouTube 频道并发布首个视频

    Descript 发布指南,介绍从零开始创建 YouTube 频道的 7 个步骤:规划频道定位、搭建频道、选择拍摄设备、布置家庭工作室、制作首个视频、优化视频获取更多观看以及频道变现。指南指出频道搭建约需 20 分钟,但真正的投入在于按固定节奏持续发布,其中定位、更新频率和缩略图一致性最为关键,Descript 负责剪辑、字幕和切片。

    g: Once you're satisfied with your video, you can export it directly to YouTube from Descript, making your workflow even more efficient Good editing takes practice. Don't be discouraged if your first few videos aren't perfect. As you gain more experience and familiarity with your editing software, your skills will improve. Keep experimenting, learning, and refining your craft. ‍ 6. Optimize Your Videos for More YouTube Views After you've put in the hard work to script, film, and edit your video,

    匹配字段正文

  17. Descript 官方产品更新

    如何从零开始做播客:面向新手的 12 步指南

    Descript 发布面向新手的播客入门指南,把开播拆成选题、形式、设备、托管平台和更新频率五个决定,并细化为 12 个步骤,涵盖受众调研、命名、封面与主题音乐、录制、剪辑和推广。文中提到美国约 31% 的人每周收听播客,并称多数播客在第十集前就停更。

    to check out include: Podbean; Acast; Captivate; Buzzsprout; Spotify for Podcasters (formerly known as Anchor); RSS.com Each has a slightly different set of plans depending on storage size and additional features. Your hosting provider will create a podcast RSS feed when you create your show, which you'll use to submit your show to whatever directories you want to appear in. Some hosting sites have built-in integrations to automatically show up on a number of platforms. 9. Write a podcast scrip

    匹配字段正文

  18. Descript 官方产品更新

    如何去除音频中的背景噪音:最快的方法

    Descript 的 Studio Sound 功能可一键去除背景噪音、混响和麦克风噪声,其机器学习模型会提取语音内容后重新合成,模拟在录音棚中录制的效果。文章还对比了 AI 降噪、噪声门和 EQ 三种方法,并介绍了录音前选择安静环境、使用合适设备等减少噪音的做法。

    cing, or reverberating, around your recording environment. It’s the difference between how you sound in a closet, where clothing or carpeting dampens the sound, versus a parking garage, where sound reflects off the hard surfaces, causing echo; Mic response. All mics pick up sound differently. A high-quality mic responds to a range of frequencies, and stays mostly true to your room’s recording conditions. But lesser-quality equipment, like built-in laptop mics, tend to hear sound differently; the

    匹配字段正文

  19. Descript 官方产品更新

    Descript 在旧金山 MUNI 车站投放首块广告牌:一篇 500 字文章登上站柱

    Descript 在旧金山 Montgomery Street MUNI 车站投放了公司首块广告牌,把一篇 500 字、讲述视频创作意义的文章放上站柱。广告视觉刻意采用反乌托邦风格,主区域强调企业需要制作更多视频,而车站角落一根柱子则表达团队最初投身视频创作的初心。

    y're impossible to satirize, just like you can't write a satirical LinkedIn post because nothing you say will be more unhinged than what 15 lunatics just posted earnestly.) The outrageous imagery is supposed to make you feel uncomfortable because the message feels uncomfortable. In creative work these days, there’s a constant tension between volume and quality. You spend a lot of time drawing the line between making good work and making enough of it to satisfy the platforms, help the business, e

    匹配字段正文

星期五

  1. The Rundown AI

    Apple 将触控带到 Mac:触屏 MacBook Pro 与智能家居新品将至

    Apple 将于 10 月 13 日在纽约举办智能家居活动,并据 Bloomberg 报道在 10 月 27 日前后推出首款触屏 MacBook Pro 和 OLED iPad mini。触屏 MacBook 预计有 14 英寸和 16 英寸两种尺寸,采用 OLED 屏幕并以 Dynamic Island 取代刘海;iPad mini 则升级为 OLED 和 A20 Pro 芯片。此外,Amazon 停产 2011 年推出的 Fire 平板品牌,改为三款内置 Alexa+ 和完整 Google Play 的 Alexa 平板,起售价 229.99 美元,10 月 14 日起在美发货。

    le, so SpaceX can sell phone service without a carrier partner; SpaceX has agreed to acquire Grain Management’s nationwide 800 MHz licenses — low-band spectrum that would push coverage indoors; It says some approaches came within tens of meters, and only half of the 650 maneuvering satellites that neared Starlink this year were sharing trajectory data; SpaceX expects its first next-gen Starlink Mobile satellites to launch by Nov. 2027, when the second phase of its EchoStar spectrum deal is set t

    匹配字段正文