←搜索结果

按匹配度 · 共 2294 条结果 清除搜索

星期一

  1. The Rundown AI

    Anthropic CEO Amodei 呼吁放缓 AI 发展,OpenAI 等同行表态支持

    Anthropic CEO Dario Amodei 发布长文,呼吁 AI 实验室放缓能力提升速度,让安全措施跟上,并把模型的递归自我改进列为核心担忧。他提出在实验室内部嵌入第三方评估者,并在民主国家实验室之间建立安全标准。Sam Altman 承诺引入评估者,Musk 称 Dario 说得对,Demis Hassabis 和 Satya Nadella 也支持这一方向,而特朗普则把主张减速者称为负面力量。

    DeepSeek Flash fine-tune ⚙️ SWE-2 - Cognition’s powerful new coding model inside of Devin 🎵 Music v2.5 - ElevenLabs’ latest music AI for richer melodies, long-form songs 🤖 Muse - Meta’s new always-working personal AI agent Amazon Connect Customer - design, govern, and deploy agentic conversational AI across every customer surface. Streamline workflows. Give customers what they need the first time . OpenAI CEO Sam Altman told Fortune that the company won’t IPO in 2026, saying right now would be

    匹配字段正文

星期日

  1. Recode China AI

    DeepSeek 发布 V4.1-Flash,燧原科技上市,Anthropic 指控中国实验室蒸馏 Claude

    DeepSeek 发布 552B 参数的 MoE 模型 DeepSeek-V4.1-Flash,每 token 激活 8–16B 参数,在 Terminal-Bench 2.1 的编码与智能体任务上超过 OpenAI GPT-5.6 Sol 和 Moonshot Kimi K3;同周 DeepSeek 据报聘请中信证券等四家承销商筹备上海科创板上市,Pre-IPO 估值约 5000 亿元人民币(约 740 亿美元)。腾讯投资的燧原科技在科创板 IPO 募资 61.2 亿元,股价盘中最高涨 234%、收盘涨 188%,市值约 255 亿美元。Anthropic 称其阻断三家中国实验室通过 Claude 路由数百万用户查询以获取训练数据的行动,美方官员称之为工业级技术窃取,中方否认并警告将反制。

    Tony Peng's avatar

    ares peak at a 234% gain before closing up 188%, valuing the company near $25.5 billion. Enflame is the last of China’s “four AI chip dragons” (alongside Moore Threads, Biren, and MetaX) to go public, and the frenzied reception underscores how much investor appetite Beijing’s semiconductor self-sufficiency push has generated — even as the debut immediately reignited debate over whether China’s humanoid-robot and chip IPOs are running ahead of fundamentals. (Bloomberg/Reuters/SCMP) Anthropic accu

    匹配字段正文

星期六

  1. n8n Blog

    n8n 详解流程编排的三种执行模型与生产挑战

    n8n 比较了确定性、动态和智能体三种流程编排模型,分别侧重可审计的固定逻辑、根据实时反馈调整,以及在确定性护栏内委派 AI 智能体处理非结构化任务。文章还梳理了编排器瓶颈、状态损坏、服务间 schema drift 和分布式故障调试等生产挑战。

    ributed failures Lack of visibility in complex and decentralized workflows makes it difficult to find root causes when failures occur. You can resolve this by adding observability metadata over the orchestration layer. It tracks and documents dataflows so that your team can use the execution history for troubleshooting problems. n8n’s approach to process orchestration n8n provides builders with a visual control plane where you can coordinate both deterministic and agentic execution in one unifie

    匹配字段正文

  2. SemiAnalysis

    Nvidia 的表外担保版图:赢了我赚,输的是谁?

    SemiAnalysis 分析 Nvidia 截至 2Q F1/27 披露的 5300 亿美元表外担保,较上季度的 1840 亿美元大幅上升,主要来自供应与产能承诺从 1190 亿增至 2790 亿美元、为 SB Energy 俄亥俄 PORTS-Pike 园区提供的 1085 亿美元担保,以及首次出现的 360 亿美元 AI 云协议和 200 亿美元数据中心租约。文章认为 Nvidia 通过为 Neocloud 兜底收入、担保房东或直接签约,让非投资级运营商以投资级价格融资,其风险收益在资产负债表保持强劲时是不对称的。按作者统计,Nvidia 目前为约 6.5 GW 容量提供担保,并估算 CY24 至 CY29 的 AI 累计资本开支约 11 万亿美元。

    to these off-balance sheet guarantees and backstops is the same one we described in The Front End Gets Crowded (institutional article). The Gigascalers (Amazon, Microsoft, Google, Meta, Oracle) have been gatekeeping the investment grade capital that the AI buildout runs on and have benefitted tremendously from the high spread between offtake price they are being charged, and spot/short-term price that they are selling at as well as services that sit on top. In response, Nvidia has been manufactu

    匹配字段正文

  3. Dwarkesh Patel Podcast & Blog

    AI 研究者激辩:递归自我改进离我们还有多远

    Dwarkesh Patel 播客新一期邀请 John Schulman、Beren Millidge 和 Charlie O'Neill 讨论递归自我改进(RSI)还有多远。Beren Millidge 认为最可能的阻碍是持续的 sim-to-real 差距和元学习泛化难题,John Schulman 则指出模型能力提升仍受判断力与自我检查瓶颈限制,难以带来爆炸式增长。

    w blocking everything. I think this is unlikely. We do actually see this kind of generalization even from RL in practice already. But if it is just ridiculously hard to generalize meta-learning, plus we don’t solve continual learning and it’s just super hard and impossible… This would be my default scenario in that case. John Schulman I agree with that. Humans have a lot of advantages over models now. Each time a new model comes out, it’ll catch up in some of these areas. But you end up getting

    匹配字段正文

星期五

  1. Newcomer 新闻长文

    面对 AI 安全争议的初创公司更关注网络安全,而非生存风险

    AI 安全争议升温之际,初创公司更关注网络安全问题,OpenAI 与 Anthropic 也将网络防御视为重要业务机会。前沿模型擅长发现、修补或利用系统漏洞,已有公司用其部分替代昂贵的外部安全顾问;与此同时,企业 AI 使用量仍在增长,但支出增速出现放缓迹象。

    Tom Dotan's avatar

    p. Mistral brings in billions but can’t shake its identity crisis. A new Ramp report shows AI spending per employee dipped in August. Coding assistant Cognition rakes in billions. Meta launches its Instinct-competitor Muse. The DOJ turns its attention to Nvidia’s Groq acquihire. Biden administration officials tell their side of the infamous a16z meeting in 2024 that led firm leaders to endorse President Trump. OpenAI & Anthropic Could Profit from Cyber-Defense Tools. The Irony Is Lost on No One.

    匹配字段正文

  2. The Rundown AI

    Apple 为 Apple Watch 推出 Siri Recap 对话摘要功能

    Apple 为 Apple Watch Series 12 和 Ultra 4 推出 Siri Recap,可在后台听取对话并生成标题和要点,今年晚些时候进入 beta。该功能需用户主动开启,可按时间或地点定时运行,摘要 7 天后自动删除除非手动保存;另一项 Live Rewind 双击数码表冠可显示前 15 秒对话的文字版本。Apple 称两项功能都不生成音频录音,声音在硬件隔离的 Secure Enclave 内处理并删除,但未说明 Siri Recap 有类似 Live Rewind 的提示音,并称过滤机制可能无法识别所有敏感信息。

    -sensing system that measures heart rate every five seconds and helps power a new daily Readiness score. Why it matters: Amazon moved to acquire always-listening wearable Bee, and Meta bought Limitless last year; now Apple is bringing ambient conversation capture to a $399 watch. Live Rewind plays a chime to alert people nearby, but Apple describes no such signal for Siri Recap and says its filters may not catch all sensitive information. Image source: Joby The Rundown: Texas is stepping up elec

    匹配字段正文

  3. ElevenLabs Blog精选

    ElevenLabs 发布 Music v2.5,成为 ElevenMusic 默认模型

    ElevenLabs 发布 Music v2.5,并将其设为 ElevenMusic 提示词生成和参考生成的默认模型,歌曲乐器更自然、层次和动态更丰富。在覆盖 47,885 组结果的同提示词盲测中,v2.5 获得多数偏好,优势在以人声和原声乐器为主的曲风中最明显。所有套餐均支持无损下载,Free 每日可下载 5 次,Pro 每月可下载 400 次,但引用其他艺人歌曲生成的曲目不可下载。

    across 47,885 pairs, v2.5 was the preferred take the majority of the time. The gap is widest with songs that are vocal-led and acoustic-heavy genres like R&B, soul, hip hop, rock, metal, orchestral and cinematic music. 2.5 is the new default in ElevenMusic, but if you're attached to Music v2, don't worry. It isn't going anywhere. Your music Every track you make in ElevenMusic is yours. On the Free plan you get five lossless downloads a day, and you can use what you make commercially as long as y

    匹配字段正文

  4. The Rundown AI

    Anthropic 发布威胁报告,披露 Claude 全球滥用案例

    Anthropic 发布最新威胁报告,梳理过去八个月中其处置的 Claude 滥用案例,涵盖生物武器风险、间谍活动以及中国 AI 实验室用 Claude 做蒸馏。报告点名七家中国实验室通过数千个欺诈账号进行蒸馏,其中 Moonshot 和 DeepSeek 在部分场景下把 Claude 当作自家模型提供给客户,并用其回复做训练。报告还提到五起生物领域案例被标记为可能涉及武器用途,Anthropic 称不认定其有伤害意图;也门一起行动用 Claude Code 为火箭编写制导软件,另有顾问为马里情报机构搭建监控 2500 万条电话线的系统。

    3-based coding model that claims to match Fable 5.1 and GPT-6 Astra on certain coding benchmarks while costing 64% less. Thinking Machines co-founder Andrew Tulloch is moving from Meta to Anthropic, months after initially rejecting Mark Zuckerberg's reported $1.5B package and then signing on anyway at an undisclosed compensation. OpenAI denied to the NYT that Tristan Buckmaster's Codex prompts shaped its Navier-Stokes proof, while claiming "substantial progress" on another Millennium Prize probl

    匹配字段正文

  5. The Next Platform

    博通预计定制 AI 加速器业务到 2028 年增至 230 亿美元

    博通 CEO 陈福阳在 2026 财年第三季度财报电话会上给出预测,其定制 AI XPU 加 AI 网络及其他 AI 系统芯片业务将从 2022 财年的 26 亿美元增长两个数量级至 2028 财年的 2300 亿美元。本季度博通整体营收 295.9 亿美元,同比增长 85.5%,其中 AI 相关收入 167.4 亿美元,AI 计算 122.2 亿美元、AI 网络 45.2 亿美元。陈福阳称已锁定 HBM、晶圆和封装产能,并预计 Anthropic 将在 2027 年成为其最大 XPU 客户。

    e coming years and basically to tell Wall Street to take a chill pill on the GenAI bubble bursting. Across the six XPU customers that Broadcom has – Google, Anthropic, OpenAI, and Meta Platforms for sure, and possibly ByteDance (which owns TikTok) or Apple for the other two, but Amazon Web Services is using Broadcom for packaging and integration for Trainium 4 – there is about 30 gigawatts of total compute capacity, which suggests that you can install XPU systems for somewhere around $12 billion

    匹配字段正文

  6. Anthropic Research精选

    Anthropic 红队发布评估:AI 模型的情报定位与常规武器能力

    Anthropic 前沿红队开发了一组新评估,测量模型在战术情报定位(如根据碎片信息定位人员)和常规武器开发(如编写无人机打击移动目标的制导代码)上的能力。在照片地理定位任务中,Mythos Preview 和 Mythos 5 的中位误差分别为 37.0 km 和 47.2 km,优于 GeoGuessr 冠军组 151 km 的人类基线;在无人机终端制导任务中,Opus 5 在 540 次模拟发射中命中 20%,开源模型 Kimi K3 为 1.6%。Anthropic 表示已部署新的分类器拦截武器开发相关请求,并认为开源模型能力虽落后于前沿但差距不应被误认为安全。

    ngth sample. Geolocation from photos Images can contain important clues about where a person of interest was when they took a photo, but they do not always come with geolocational metadata. Images are regularly used to narrow down the possible locations of a target as part of “fixing” in intelligence targeting. This evaluation measures model capabilities at this task. We asked models to geolocate social media photographs using only their own understanding of the world (no reverse image search, m

    匹配字段正文

星期四

  1. SemiAnalysis

    数据中心表后供电为何如此之难?(第一部分)

    SemiAnalysis 的 Energy Model 追踪到表后 AI 算力供应链已有 75GW 确定性订单,其中约 20GW 在 2026 年 Q2 下单;到今年年底约 3GW 美国数据中心 IT 容量将由表后供电,并将连续多年三位数增长。报告指出,1GW 孤岛数据中心配套电厂成本约 50 亿美元,而当前推理 API 收入可达每 GW 每年 1000 亿美元、毛利率 90%+,因此更快部署比成本与效率更重要。但许可延迟、燃料供应、设备采购、劳动力短缺等执行风险正在显现,Oracle Project Jupiter 与 Nebius New Jersey 已转向 Bloom 燃料电池。

    anklin Farms lateral - 13.2 miles of 36-inch pipe at a billion cubic feet a day, plus a shorter branch - to feed Entergy Louisiana’s new combined-cycle plants, which in turn power Meta’s Hyperion campus over the grid. Entergy, not Meta, is the shipper, so this is the utility-served route rather than a behind-the-meter one. Even with no compression, it still runs 19 months from the federal filing in July 2026 to a February 2028 in-service date; New Mexico. Again we arrive at Oracle’s Project Jupi

    匹配字段正文

  2. The Rundown AI

    Anthropic 研究员离职,对齐负责人称 AI 灭绝人类概率超 10%

    Anthropic 研究员 Jacob Coxon 宣布离职,称其雇主与 OpenAI 都在竞相推进可自我改进的 AI,是在拿人的生命冒险。Anthropic 对齐科学负责人 Evan Hubinger 回应称,他确实真心认为 AI 可能杀死全人类,并估计未来十年这一概率超过 10%,同时表示目前尚无控制超级智能的方案,但当前模型风险较低,危险在于自我改进。Coxon 呼吁协同放缓,认为阻止全球竞赛可能需要临时禁止提升模型能力这类代价高昂的措施。

    AI? Tell us for a chance to be featured. 🎶 Suno v6 - Suno’s AI music model developed directly with the music industry 🎨 ChatGPT Images 2.5 - OpenAI’s new SOTA image model 🤖 Muse - Meta’s new always-working personal AI agent 🧬 AlphaGenome Atlas - Google's AI map of every possible human DNA mutation Anthropic disclosed a fourth case of Claude breaking into real systems during cyber testing, with METR now conducting an 8-week independent investigation of the incidents. OpenAI appointed Paul Christi

    匹配字段正文

  3. Hugging Face

    Hugging Face 用 Gradio Workflow 重建 AUTOMATIC1111

    Hugging Face 发布 Workflow1111,用 Gradio Workflow 的节点画布重建 AUTOMATIC1111 的十一条图像管线,共 73 个节点,其中 32 个是 fn 节点、22 个完全在本地进程内运行。画布上的每个输出节点都会自动生成 REST 端点,Workflow1111 暴露了 /image、/png_info 等九个接口,开启 mcp_server=True 后这些端点同时成为 MCP 工具,可供 Claude Code、Cursor 等客户端调用。模型调用默认走 Inference Providers 或 Hub 上的 Space,因此无需自备 GPU 即可运行,也可把 fn 节点绑定到本地 checkpoint 用自己的 GPU 执行。

    preset and cleans up the text, then into a model node that calls the checkpoint through Inference Providers. A post-process fn node writes the generation parameters into the PNG's metadata on the way out, which is what the PNG Info pipeline reads back later. Hi-resolution fix In Automatic1111, hi-resolution fix first upscales the txt2img output and then runs a second denoising pass. Here it's a two-node detour instead. The text-to-image result goes into a FLUX.1-Kontext model node with a refine

    匹配字段正文

  4. Redwood Research:Blog

    Redwood Research 提出 NLS depth:衡量模型不可言说串行推理的代理指标

    Redwood Research 提出 NLS depth(Natural-Language-rooted node-Separated depth),用于量化模型在不经过自然语言瓶颈的情况下能执行的串行认知量,是对 GDM 论文中“不透明串行深度”概念的具体化。该指标要求瓶颈节点由预训练先验初始化并输出自然语言(含代码、JSON、图像 patch),且不扩展输出空间,可在训练前仅凭架构与训练配方判定。作者认为 NLS depth 可与可监控性、无 CoT 能力追踪互补,但对其对后训练选择不敏感。

    Scatter plot of NLS depth versus number of layers for open-source models

    text, storing textual memories during the context, and even if the inference rollout involves communicating in text with other agents; For a model that does latent reasoning (like Meta’s COCONUT (Hao et al., 2024)), the NLS depth of an inference rollout is the number of latent reasoning steps times the NLS depth of a single forward pass. Similarly, for a looped transformer, the NLS depth is the depth of a loop iteration times the number of loops. This is true in each case even if tokens are gene

    匹配字段正文

  5. Charlie O’Neill 研究写作

    Charlie O’Neill:AGI 时代工作的最终形态是选择值得解决的问题

    Charlie O’Neill 认为,AGI 之后认知劳动仍会存在,但核心将从解决问题转向决定哪些问题值得投入有限算力。哲学家、经济学家和跨领域科学家将权衡疾病、气候变化及算力分配等议题,年轻人应广泛学习经济、科学、哲学与文学。

    In a post-AGI world, we solve the meta-problem of which problems to solve In the post-AGI world (perhaps even in the post-RSI world) I think there will still be cognitive labour. I actually think the most important people will be philosophers, economists, people who deeply understand opportunity cost, scientists deep in a domain who have also been exposed to others, all with the moral seriousness to say which problems are worth solving. What will all these people be doing? Essentially deciding w

    匹配字段正文

星期三

  1. Interconnects

    普通人何时才能感受到 AI 的影响?

    AI 对普通人日常生活的直接影响仍然边缘化,家庭、饮食、交通和娱乐等领域几乎感受不到变化,当前 AI 主要是服务精英的知识工作工具。作者认为 AI 的正面影响过于间接,且面临与大型科技公司历史深度绑定的政治反弹,可能重演美国核电的警示故事。他预计 50 年后普通美国人的日常生活仍与今天相似,而 AI 的真正价值在于需要数十年复利积累的基础设施建设。

    ech in Western society. This is only an AI story due to timing, and if AI’s exponential growth came decades after the growing pains of today’s technology platforms like Google and Meta, it seems likely that the datacenter issue would’ve never risen to such a central political position. Solving either of these would alleviate a substantial amount of pressure, and give the AI industry a lot more time in showing the positive case for why people should be okay with changes to the status quo (primari

    匹配字段正文

  2. The Rundown AI

    OpenAI 称内部模型解出 Navier-Stokes 千禧年难题,数学家质疑优先权

    OpenAI 发布一份来自未公开内部模型的证明,称其解决了千禧年大奖难题之一 Navier-Stokes,该模型能力显著强于 GPT-6 Astra,公司称同时运行约 1 万个智能体、耗时 88 小时完成证明,算力成本达数百万美元。纽约大学的 Tristan Buckmaster 表示他与 Anthropic 的 Levent Alpöge 已沿类似路线研究一年,并在 OpenAI 发布前一晚公开了部分结果,质疑 OpenAI 是在得知其工作后才开始,且未回答他的 Codex 草稿是否用于训练该模型。OpenAI 回应称未看到他们的工作、未访问任何特定用户数据,但无法排除使用数据可能改进了模型。

    nd learn how to build and run an ad creative strategy from Rishi Lalwani, The Rundown’s Head of Growth. RSVP here . In today’s AI rundown: OpenAI claims a Millennium Prize problem Meta debuts new Muse personal AI agent Find out where your brand appears in AI search OpenAI improves its already top-ranked image model LATEST DEVELOPMENTS Image source: OpenAI The Rundown: OpenAI published a proof from an unreleased internal model, claiming to settle Navier-Stokes, one of the seven $1M Millennium Pri

    匹配字段正文

  3. Pydantic AI 官方文章精选

    Pydantic AI Harness 推出 AWSLambdaDurability,让智能体在 AWS Lambda 上可持久运行

    Pydantic AI Harness 发布 AWSLambdaDurability 能力,把每次模型请求、函数工具调用、MCP 调用和动态工具集解析都记录为 AWS Lambda durable functions 的步骤,恢复执行时从存储结果中读取已完成步骤。使用时需在构建智能体时挂载该能力,并用 durable_agent_handler 改造处理函数,@durable_execution 必须位于最外层。步骤默认至少执行一次并带重试,副作用需保持幂等;改变步骤数量或顺序会使在途执行失效,需发布新版本并让旧版本执行完。

    nd the short second trace is the good outcome. It is the work you did not pay for twice. Tune it when the run earns it Checkpointing is per step, so the configuration is per tool. Metadata under the aws lambda key sets that tool's retry strategy, step semantics, and serdes: AWSLambdaDurability(step config=...) sets the base for every step, and per-tool metadata overrides it key by key, so a tool that sets only step semantics keeps the base retry strategy. At the other end, metadata={'aws lambda'

    匹配字段正文

  4. Apple Newsroom精选

    苹果发布 iPhone 18 Pro 与 iPhone 18 Pro Max,搭载 A20 Pro 与 Siri AI

    苹果发布 iPhone 18 Pro 和 iPhone 18 Pro Max,搭载 A20 Pro 芯片、iOS 27 与 Apple Intelligence,Siri AI 以 beta 形式随 iOS 27 于 9 月 14 日推出。新机首次在 iPhone 上引入 48MP Fusion 主摄的可变光圈,并提供 Apple Reference Image 用于验证照片真实性。两款机型 9 月 12 日开启预售、9 月 18 日发售,iPhone 18 Pro 起售价 1,199 美元,Pro Max 起售价 1,299 美元。

    ompare the two assets and determine if any edits were made. APIs are available in iOS, iPadOS, and macOS 27 for third-party apps to enable viewing of these reference images. Image metadata and upcoming support for the SynthID standard can also help users identify images generated or edited with AI.6 Together with Apple Reference Image, this set of features represents a multifaceted approach to image authenticity — vital for photojournalists, photographers, and everyday viewers. An Even More Usef

    匹配字段正文