X（推特）上的财经、科技与宏观账号
Great read 引用 @arpitrage：I spent a couple of days at Anthropic learning from their folks about AI advances, and here's what I took away: • We are close to RSI (ie, models building other models) which may lead to rapid takeoff scenarios (with risks and benefits). Some evidence for this is code per
SynthID入选《时代》2026年最佳发明之一。已有超过1800亿张图片和视频添加了SynthID水印，帮助人们区分AI生成内容与其他媒体。
StepFun宣布Step 5 Preview开放权重，采用总参数量600B、每个token激活27B的稀疏MoE架构，支持1M-token上下文窗口和原生文本、视觉输入。
推荐理由：StepFun公布Step 5 Preview开放权重，并列出模型规模、上下文窗口与输入形式，NVIDIA则补充了GB300上的运行速度和NVFP4口径。
Nemotron 3.5 Super VL 将文本和视觉理解整合到一个模型中，减少在独立文本模型与视觉模型之间传递上下文的需要。აგენტ可能需要阅读文档、解读图表并搜索网页来完成任务。
英伟达推出开放模型 Nemotron 3.5 Super VL，将推理和工具使用与文本、图像、文档及视频理解结合，用于长时间运行的 AI agents。该模型总参数量为 120B，激活参数量为 12B。
Muse Spark Open 今日可通过 Baseten、Fireworks AI、Groq、Modal、Nebius、OpenRouter 和 Together AI 的 API 平台使用。
Meta发布开放权重模型Muse Spark Open，称其为迄今能力最强的美国开放权重模型，面向编程和智能体任务。模型总参数量为816B、激活参数量为42B，支持1M token上下文窗口；NVIDIA称该模型在NVIDIA GB300 NVL72上训练。Meta还制作了针对Blackwell Ultra优化的NVFP4检查点，原文随后提到“over 50K”，但未提供完整表述。
英伟达宣布扩展 Nemotron 开放模型家族，加入 Nemotron 3.5 Super VL。该模型结合视觉理解、推理与工具使用，面向 AI agents；引用内容称其支持文本、图像、文档和视频理解，总参数量为 120B、激活参数量为 12B。开放模型可供用户定制、部署和构建。
GPT-Rosalind is built to support that work: helping researchers reason across scientific literature and biological data, with scientists bringing the expertise and judgment to evaluate the evidence. Eligible labs can receive up to 12 months of sponsored access to OpenAI tools,
Working out how best to use AI in mathematics research—and in science more generally—will take time, effort, and resources. We’re eager to support mathematicians as they drive scientific progress and define the future of their field. We’ve committed $5 million of funding, 引用 @SimonsFdn：How can the
NemoClaw was just named one of TIME’s Best Inventions of 2026! 🦞 So proud of the team and shoutout to everyone building safer AI agents with us. Blurb on @TIME: https://t.co/EY5S9htCn5 https://t.co/G3vo2N3RPP 引用 @TIME：These are creations changing how we live, work, play, and think about what’s p
Grok @Bot improves significantly every day 引用 @jumperz：i was about to pay for a small Codex subscription just for computer use.. honestly, I didn't expect Grok bot's computer use to be this fast... I've been more and more impressed with it every day lately. for day to day stuff, nothing else I've
NVIDIA表示，FLUX 3 Image Klein 5.6B在NVIDIA GB200 NVL72上训练，并可在RTX GPU上运行。NVIDIA还提到该模型有NVFP4变体；引用的@bfl_ai称其为开放权重模型，拥有56亿参数，可在单张消费级GPU上运行，并支持原生4K、边界框和像素级编辑。
财经博主转述并质疑Gartner关于新云服务商的预测，询问其量化依据是否可信。引用内容称，Gartner预计到2030年80%的新云服务商将停止独立运营；2026年AI服务商预计将在AI基础设施上支出近1万亿美元，并预测2027至2030年再投入6万亿美元。
亚马逊推出两款利用人工智能记录并分析用户日常活动的可穿戴设备。此举是亚马逊重塑 Bee 的努力之一；Bee 是亚马逊去年收购的一家初创公司，亚马逊希望借此参与 AI 硬件领域竞争。
In four years ChatGPT went from a fun toy to a tool you can use to reach and then advance the frontier of knowledge in numerous subjects. But also you have to pay for it, with either money or attention. "Enshittification!" 引用 @josephfcox：OpenAI continues to enshittify ChatGPT: OpenAI allows adverti
Hello, Stockholm. 🇸🇪 From the podcasts we discover to the robots working in factories, the Nordics are building with NVIDIA. Spotify is making podcast recommendations more efficient. ABB Robotics is helping engineers train and test robots in simulation. And startups are https://t.co/2OYTR0SoE6
This works well. If Grok @Bot makes a mistake with a purchase or leaks your card (never happened yet), SpaceX will make you whole. 引用 @XFreeze：Grok @bot booked a flight for me today the entire process was really really easy…..I send a photo of my credit card and it did the entire process…. Tbh,
Anthropic投资者日披露三季度营收后，市场担忧其增速放缓、年底ARR达到千亿美元难度较大，但前沿大模型厂商营收并不能代表AI商业化全貌。文章认为，企业正将AI采用推向成本工程化，高频、低复杂度任务倾向更便宜的Token、模型和云端部署，并预计三季度云厂商财报会大超预期，理由包括二季度指引强、Agent热潮及企业采用加速。
Anthropic第三季度ARR基本持平；若7月26日ARR为650亿美元、第三季度营收年化为672亿美元，则几乎没有ARR增长。被引推文称，Anthropic第三季度营收为168亿美元，高于第二季度的115亿美元，但季度增速从第二季度的143%放缓至第三季度的46%。
But seriously. This is why you need to be anthropomorphizing. Not because they’re sentient beings, but because anything a human might do or say — like “maybe I should hijack the compiler and be a legend” — the models will also do or say. 引用 @ArthurCDent：Apparently this is the reasoning trace from G
Is there something other than this tweet tanking the AI stocks today? 引用 @elonmusk：@chamath Yikes
Anthropic在向投资者的闭门沟通中披露第三季度财务情况，营收为168亿美元，较第二季度的115亿美元增长约46%。正文还提到公司实现调整后盈利，但相关金额和口径说明未完整呈现。
You give an AI agent a task. What happens next? This video walks through how agents use models and tools to complete tasks, and what makes them fast and responsive. No technical background needed. https://t.co/zoKBeOPqwL
To advance more research with Gemini 4 Argon, we’re offering 10,000 scientists access to our highest usage tier in @GoogleAntigravity free of charge for 6 months. Find out more → https://t.co/A9F2OZi4OJ
Argon also compressed the process of enzyme engineering into a matter of hours to design five lab-verified serine esterases. 🧬 https://t.co/74xPM9LJPa 引用 @RyanKrue：Over the past few weeks, we have been putting Gemini 4 Argon to the test on protein engineering workflows at Google DeepMind, and we h
Try Grok Voice 引用 @SpaceXSI：Introducing Grok Voice 3.0, the world’s most intelligent speech-to-speech model. It excels at resolving customer calls, responds in under a second, and costs less than other frontier models. https://t.co/cfDRrkQPEc
Grok @Bot can detect scam emails 引用 @wmorrill3：An email from noreply@x.ai landed in my inbox, passed SPF, DKIM, DMARC, and had the blue check Something felt off, so I asked @bot to do a security review. It found a hidden phishing link, explained why every safety check passed, wrote a report to xA
Watch the full convo with @finkd, @alexandr_wang and @collision: https://t.co/AJ9GHXjB2d 引用 @collision：Cheeky Pint is back! Mark Zuckerberg @finkd and Alexandr Wang @alexandr_wang came to the pub to discuss their new frontier model, Muse Wonder. 00:00:24 Muse’s breakout use cases 00:07:28 The mod
Grok is the best conversational voice model in the world 引用 @SpaceXSI：Introducing Grok Voice 3.0, the world’s most intelligent speech-to-speech model. It excels at resolving customer calls, responds in under a second, and costs less than other frontier models. https://t.co/cfDRrkQPEc
OpenAI 招募了一名负责基础设施和算力业务的 xAI 高管，Sachin 在任职一年后离开，并表示将在几周后开始新的事业，方向仍与算力有关。他称，过去一年已与合作伙伴在俄亥俄州、朴茨茅斯、佐治亚州、Trainium 等项目中落实了两位数 GW 的算力。
To emphasize, look at B200 indices on BBG for Compute Desk, Ornn, and Silicon Data. Apples to oranges as compute desk is 1-year reserved/committed pricing vs the other two are more levered to real-time recently struck rates or spot rates. Interesting the movement of each, though. https://t.co/2VY4yq
一则评论认为，Anthropic全部“有约束力”的基础设施相关承诺，预计对应2027年560亿美元、2028年860亿美元的算力支出。评论称，相比没有时间考量、也未区分固定付款与里程碑付款的5000亿美元总额，这一估算没那么令人担忧。
Made with @Grok @Imagine 引用 @keepgoingAnnie：For those who didn't want the headache here is the same video, but removing the added effects so the light flashing isn't happening anymore or the harsh coloration. I hope this version is more enjoyable for those who share the same problem. Thanks for let
Grok @Bot can help with your taxes 引用 @jasonfreedman：I just gave my a Grokbot four years of past tax returns and financial files. Just a dump of folders and unlabled files. K-1s, 1099s, mortgage statements, HSA records, and every email with our CPA. 266 files. I just said, please double check m
Grok @Bot can help with complex taxes 引用 @Texas_Twins_Dad：Grok Bot saved me more than $1,000 today. I had a very complex tax filing with overseas reporting. Complicated, expensive, and hard to find the right accountant. Grok Bot found and organized the necessary documents itself (emails and bank a
新加坡超过一半的报案犯罪使用了复杂的社会工程手段。 我们正与 @GovTechSG 合作，使用我们的开放模型 Gemma 构建端侧欺诈检测。 通过针对本地诈骗手法进行微调，它能实时识别威胁，同时不损害个人隐私 → https://t.co/VdCio2nrHC 原文： Over half of reported crimes in Singapore use sophisticated social engineering. With @GovTechSG, we’re building on-device fraud detection using our open model Ge
法国总统马克龙表示，英伟达将投资法国企业，并承诺支持法德发展欧洲开放研究的倡议。马克龙称，自“选择法国”（Choose France）以来，双方合作扩展至初创企业、大型工业企业、计算基础设施和研究，并共同强调开放、安全、服务于主权的人工智能。
开源模型若能提供足够价值，推理服务商即使降低 token 价格，仍可在算力成本之上保留利差。被引评论提出，这种模式结构上可能利好算力需求，但短期可能压低算力价格；评论还提到 AI 实验室每 GW 收入可能最高。
这和往常一样是一串精彩的评论，但我主要的一个问题是：这在结构上会利好算力需求，但短期内利空算力价格，对吧？ 我猜如今 AI 实验室是每 GW 创造收入最多的。如果我以 Ant 作为 https://t.co/Ei2VtLpy9U 被引原文（@GavinSBaker）：1) 再试一次。 TLDR：每个人都能赢。开源、前沿实验室、基础设施、应用，最重要的是美国人民。 我理解所有关于 Anthropic 丢失份额、实际利率、利差和二阶导数的紧张情绪。仍然认为最可能的 https://t.co/PeSOvvC706 原文： This is a brilliant thread of comme