Let’s get building, Paris. 🦞
Let’s get building, Paris. 🦞 @JensenHuang stopped by Build-A-Claw, where developers and startups are getting hands-on with open models to build AI agents of their own. https://t.co/ipezGKUz0X
#人工智能
X（推特）上的财经、科技与宏观账号
X·科技公司AI 评分1919 X·科技公司AI 评分1010 Here’s how you can try it:
Here’s how you can try it: 🔵 Everyday users: In the @GeminiApp and AI Mode in Search for Google AI Pro and Ultra subscribers. 🔵 Developers: Gemini API paid tier, plus @Antigravity and @GoogleAIStudio for Google AI Ultra subscribers. 🔵 Scientists: Apply for the top usage tier https://t.co/4IH6A984
X·科技公司AI 评分4242 Gemini 4 Argon 现已全球可用
Gemini 4 Argon 现已全球可用。该模型面向从头到尾解决复杂问题，具备编码、企业工作流和网络防御方面的前沿能力。
X·科技公司AI 评分44 Customizing Vertical Voice Agents with ASR and LLM Fine-Tuning
Customizing Vertical Voice Agents with ASR and LLM Fine-Tuning https://t.co/dBSfkUiKeX
X·科技公司AI 评分2929 Congrats! CEOs & execs posting to the public directly on 𝕏 is the best way to get your message out…
Congrats! CEOs & execs posting to the public directly on 𝕏 is the best way to get your message out and drive product sales. Vastly more powerful and effective than a press release. 引用 @sundarpichai：Gemini 4 Argon is here. It’s available globally today for developers and enterprises in the paid
X·科技公司AI 评分3838 . wrapped up his Paris visit at the Élysée with President , discussing NVIDIA’s investments in…
.@JensenHuang wrapped up his Paris visit at the Élysée with President @EmmanuelMacron, discussing NVIDIA’s investments in France’s AI ecosystem, open source, and AI safety. Our road to #NVIDIAGTC Berlin continues. https://t.co/CGYw6B8UFl
X·科技公司AI 评分1818 All Grok get their own email address and can help analyze and summarize your email box too
All Grok @Bots get their own email address and can help analyze and summarize your email box too 引用 @ericzakariasson：grok @bot 0.70.2 is rolling out! - each bot can claim its own email address at @mail.grokbot.com and draft emails for you to send - apps like Figma, Notion and Canva show up as live
X·财经博主AI 评分4949 英伟达黄仁勋称一周内将宣布20亿美元欧洲投资，并点名四家法国初创公司
英伟达黄仁勋表示，将在一周内宣布20亿美元的欧洲投资，过去一年已投资80亿美元。黄仁勋在巴黎点名四家法国初创公司；英伟达今年前四个月承诺投入超过400亿美元用于AI股权投资。
X·科技公司AI 评分1818 To quantify how the models handle unfamiliar tasks, we compared test-time compute scaling on tasks…
To quantify how the models handle unfamiliar tasks, we compared test-time compute scaling on tasks trained on during post-training against other tasks. Both models improve similarly out-of-distribution, but given the same in-distribution score, Muse Wonder Full performs better https://t.co/SES4eMwL
X·科技公司AI 评分44 Two things make Muse Wonder suited for long-running, multi-step work an agent faces every day:
Two things make Muse Wonder suited for long-running, multi-step work an agent faces every day: -> Active Memory carries your context across conversations and surfaces it when it's relevant. -> Active Focus keeps the models working toward your goal when a task gets challenging, https://t.co/4t3lIYq
X·科技公司AI 评分3434 Great to see what the Fin team has built with NVIDIA Nemotron 🙌
Great to see what the Fin team has built with NVIDIA Nemotron 🙌 They fine-tuned Nemotron 3 Super and Nano for their new Fin Apex customer service models. Congrats on the launch! 引用 @fergal_reid：We’re pleased to announce the launch of our 2.0 series of Fin Apex models! We think these are the bes
X·科技公司AI 评分2020 Over the last few months, we’ve made significant progress in reducing the effectiveness of…
Over the last few months, we’ve made significant progress in reducing the effectiveness of distillation attacks against Claude. In this post, we explain how: https://t.co/uWiRSLBuOS
X·财经博主AI 评分4343 Meta超级智能实验室成立一年多，推出Muse Wonder前沿模型家族
Meta超级智能实验室成立一年多，推出Muse Wonder前沿模型家族，面向个人和专业智能体工作，性能达到最先进水平。
X·财经博主AI 评分2020 Super interesting: "Anthropic's Claude Fable 5.1 scored 41% in Claude Code and 61% in OpenAI's…
Super interesting: "Anthropic's Claude Fable 5.1 scored 41% in Claude Code and 61% in OpenAI's Codex. It made 28 tool calls at the median in one and 66 in the other." "No model did best in its maker’s own harness" https://t.co/BONmxl9hiE 引用 @chamath：Model Benchmarks Are Worthless How can a model
X·财经博主AI 评分6666 媒体称微软自研 Maia 200 AI 芯片已投产，Maia 300 计划明年启动制造
媒体称微软自研 Maia 200 AI 芯片已投产，下一代 Maia 300 计划明年开始制造。Maia 300 的计算芯粒将由台积电采用 2nm（N2）制程制造；原文末句关于微软部署计划未完整。
X·宏观AI 评分4545 红杉两年前提出的6000亿美元问题，如今已变成3万亿美元问题
红杉两年前提出的6000亿美元问题，如今已变成3万亿美元问题。相关支出规模扩大之际，实际收入仅增加略多于1000亿美元，前瞻性需求与现实收入之间的差距已扩大。
X·宏观AI 评分3939 1/5 The S&P 500 increasingly looks like two stock markets: AI-related companies and everybody else.
1/5 The S&P 500 increasingly looks like two stock markets: AI-related companies and everybody else. 41 AI-related companies we track accounted for 48.68% of the S&P 500's combined market capitalization. The remaining companies accounted for 51.32%. https://t.co/owS3WGJiom
X·科技公司AI 评分1515 Grok can research best prices and buy things for you
Grok @Bot can research best prices and buy things for you 引用 @Teslaconomics：Bro… you still open websites, add things to your cart, and type in your credit card info just to buy something online? I just text my @bot what I want, and it handles everything. Literally finds it, checks out, places the
X·财经博主AI 评分1111 Shower thought this AM: Some people developing AI seem to view building a “digital god” or ushering…
Shower thought this AM: Some people developing AI seem to view building a “digital god” or ushering in a new species in almost religious terms. I believe some of that conviction is sincere. But I also suspect there’s a marketing angle to giving these systems an identity and
X·财经博主AI 评分4040 OpenAI 与 Anthropic 占用的 AI 算力份额关系到推理算力占比达到 90% 的说法
作者指出，AI 实验室占总算力的份额关系到推理算力的测算，并以假设的 100 单位算力为例：若 OpenAI 与 Anthropic 使用 50 单位，其中 40% 用于推理，则推理算力为 20 单位。Marvell Technology 高管 David Lazovsky 称，数据中心基础设施中用于推理的比例 3 年前为三分之一，如今超过 60%，未来 2-3 年将迅速接近 90%。
X·财经博主AI 评分1111 Seeing this more with Grok Bot as well. But even outside the personal agents, see it with the…
Seeing this more with Grok Bot as well. But even outside the personal agents, see it with the normal agents thru Chat or Desktop app w/ computer use. Annoying once you've seen the convenience of them being able to do things for you and now just clearly hamstrung. 引用 @matthew_sigel：Nearly every shop
X·宏观AI 评分1818 We actually recorded this Tuesday, just a few hours, before OpenAI's latest release. So we mostly…
We actually recorded this Tuesday, just a few hours, before OpenAI's latest release. So we mostly talked about the consequences of the Navier-Stokes earthquake. But basically, I think, you can jus take this whole conversation and now multiply it by 10.
X·财经博主AI 评分1818 Like usual growth solves most problems so there’s a chance this looks like a blip if the labs go…
Like usual growth solves most problems so there’s a chance this looks like a blip if the labs go back to rapid ARR growth This was always the issue when every bet was additional leverage on the same data points (lab growth). That’s why fear cascades when the foundational bet 引用 @joinyellowbrick：I
X·科技公司AI 评分66 Using Grok is like hiring a super smart, hard-working person for peanuts
Using Grok @Bot is like hiring a super smart, hard-working person for peanuts 引用 @B_doong2daddy：First month I paid $200 Cursor Ultra just to try Grok Bot. I said to myself.. it's okay even I don't find it worth it, I will still learn something. I learned so much since I started Grok Bot + Cursor. I
X·宏观AI 评分1717 It seems like the history of AI is basically a series of one technique augmenting an older…
It seems like the history of AI is basically a series of one technique augmenting an older technique, and then just dropping the older technique altogether. So does that eventually mean we can just have RL and no need for an original pre-training run? 引用 @not_ellington：It really does seem pre-train
X·宏观AI 评分88 If something has been proven, but nobody knows what the proof is (in language that humans can…
If something has been proven, but nobody knows what the proof is (in language that humans can understand), maybe that becomes the job of mathematicians. Though I suppose the bitter lesson is that eventually, AI also gets better than humans at translating the outputs of AI. 引用 @ValerioCapraro：BREAKI
X·宏观AI 评分3333 FT: OpenAI ARR比先前市场预估的少了200亿，市场下跌
FT: OpenAI ARR比先前市场预估的少了200亿，市场下跌 这种新闻怎么理解？ 1. 首先ARR不是营收，你可以拿1个月去年化，1周去年化，或者1天去年化，除非内部改标准（比如上年用1个月年化，今年用1天年化），那就是很严重的欺诈，不然只要标准一样没问题，看的是同比增速。当然上市后有监管的衡量方法 2.
X·宏观AI 评分1919 Expired: AI will take your job.
Expired: AI will take your job. Tired: A human working with AI will take your job. Wired: A human working with an unreleased model at an AI company will take your job. 引用 @TheStalwart：THE REAL IMPLICATION OF THE AI MATH BREAKTHROUGHS I wrote about an element of AI that needs to get more discussi
X·宏观AI 评分2020 Very consistent with the point that the physical buildout of AI/ "SI in Trump land" computer…
Very consistent with the @DKThomp point that the physical buildout of AI/ "SI in Trump land" computer infrastructure is much more visible in the actual data than the results of actual AI use 2/2
X·科技公司AI 评分1111 Grok works well on your phone. Just download the app from Apple or Android.
Grok @Bot works well on your phone. Just download the app from Apple or Android. https://t.co/Efbxq5vqTm 引用 @melvindvivas：This feels unreal Used Grok Bot today, mostly from my phone I got more done from my phone with Grok Bot than I usually do sitting at my desk - Colab notebooks for Google Em
X·财经博主AI 评分2121 You can debate whether you believe their forecast on capex, labs, etc. but Coatue always does a…
You can debate whether you believe their forecast on capex, labs, etc. but Coatue always does a nice job framing some of the biggest trends in the market very simply. The size of the compute financing & capex wave relative to the broader economy and prior cycles relative to the https://t.co/H5Yeb64
X·财经博主AI 评分5050 Broadcom与OpenAI拟融资500亿美元，Apollo和Blackstone有意参与；SpaceX拟为GPU芯片寻求400亿美元融资方案
Broadcom与OpenAI拟融资500亿美元，Apollo和Blackstone有意参与。SpaceX拟为GPU芯片寻求400亿美元融资方案，PIMCO据称也有兴趣。引用内容称，这笔交易将体现GPU融资市场的开放程度。
X·财经博主AI 评分3838 Gigabrains help me out:
Gigabrains help me out: GS puts required annual hyperscale AI revenue at $526bn, based on average annual capex across 2026 and 2027 and a 20% hyperscaler ROIC. I assume OpenAI and Anthropic have similar ARR, GW capacity and inference share, with $20bn of revenue per blended GW
X·财经博主AI 评分1414 I’d love economists to take a crack at this theory that the US is experiencing a mini-Dutch Disease…
I’d love economists to take a crack at this theory that the US is experiencing a mini-Dutch Disease effect from AI. My big-picture thesis about AI economics right now is that while the discussion about AI safety and existential risk is important, AI’s construction is changing 引用 @johnarnold：This.
X·宏观AI 评分2323 Great analysis by a high IQ person in the industry.
Great analysis by a high IQ person in the industry. 引用 @hsu_steve：AI Boom or Bust? – Manifold #122 Are we in an AI bubble? Analysis of AI CapEx, revenues, OAI+ANT burn rates. AI revenue has to reach $ TRILLIONS per annum within the next few years (an increase of > 10x; compare total global soft
X·宏观AI 评分4343 Grok Bot将按任务挑选最佳模型，不再锁定自家模型
Grok Bot未来将按具体任务挑选后端模型，不再锁定自家模型。文本推理可能使用 Claude Opus 5.5，生图可能使用 MidJourney，音频或音乐可能使用 Suno。
X·宏观AI 评分2828 Cooling LLM Spend
Cooling LLM Spend Timely data suggests paid models ARR growth slowing considerably in recent months. How will the needed $3tln/yr in revenue come by 2030 when linear growth (or worse) is happening at 120bln/yr? h/t @Silicon_Data https://t.co/WtFsXAjiSP https://t.co/34IoGdPhrU
X·宏观AI 评分1515 While the world frets human extinction from AI, and I look at AI’s impact on humans’ jobs:
While the world frets human extinction from AI, @KevRGordon and I look at AI’s impact on humans’ jobs: https://t.co/Lm87VvW5YM
X·科技公司AI 评分1717 SynthID Detector can quickly identify our watermarks in video, audio and images – even after edits…
SynthID Detector can quickly identify our watermarks in video, audio and images – even after edits, like adding filters. ✂️ You can also use existing AI identification features found in @GoogleChrome, @Google Search, and the @GeminiApp. https://t.co/VglUeteVIL
X·科技公司AI 评分3333 SynthID Detector is now available to everyone. 🌐
SynthID Detector is now available to everyone. 🌐 Check whether online content was generated using @GoogleAI, or with tools from our industry partners – including @OpenAI, @NVIDIA, Kakao and coming soon, @Apple. Try it out → https://t.co/cPW2aNnqr6 https://t.co/TGYA0uZ0dx