X（推特）上的财经、科技与宏观账号
Grok @Bot will use whatever achieves the best outcome for users. Simple questions will route to small, fast models. Questions with complex answers will route to large models. 引用 @Scobleizer：Grok goes open. Holy shit.
However, most @Bot requests are pretty simple and will be handled by a lightning-fast version of Grok 4.8 when that comes out. Operating principle is to give Grok Bot users the best possible combination of speed & intelligence. 引用 @minchoi：This is wild... Grok Bot will now pick Claude Opus 5.5,
Neovance gets 15,000 patient calls a week and AI handles 18% of them. Patients aren’t waiting in a queue, and pharmacists can focus on patient care. https://t.co/jqK0TciYuV
An AI agent makes a mistake early in a task, then keeps going in the wrong direction. Our researchers built PivotOPD to teach agents how to avoid those mistakes and recover when they happen. During training, a teacher model shows the agent a better action and how to get back on https://t.co/7hMLMdE
Grok Bot 将采用灵活的模型部署方式。SpaceX 表示，今后会针对不同任务选用最合适的后端模型，包括 Claude Opus 5.5、MidJourney、Suno 等领先 API，以提供更好的结果。
Shanu Mathew转述，谷歌目标是每六个月将有效 token 服务能力翻倍，而非仅提升原始 GW、FLOPs 等指标。引用内容称，谷歌 AI 基础设施首席技术专家 Amin Vahdat 在 Transition-AI 2026 活动中谈到算力容量每六个月翻倍；原帖还提及 Training Data podcast 和“Training → Inference cascade”。
作者称，黄仁勋认为MRVL未来达到万亿美元市值的概率较高。引用内容称，MRVL宣布与谷歌于7月29日达成芯片产品开发协议，涵盖AI、存储和网络芯片领域，协议潜在总值为1200亿美元；MRVL还向谷歌发行最多可购买5897万股股票的认股权证，行权价格为每股206.58美元。协议签署后的第一年内，每季度将有136.0867万份认股权证股份归属。
$GOOG 收购 $CEG，不是在卖飞就是在卖飞的路上，等着 $VST 弥补我的损失！ 能源 = 电 = 数据中心 = TOKEN = 智力 = 营收
What bothers me about this Krugman piece is that he's still doing the "AI doesn't work" thing from the 2024 era progressive epistemic bubble. Progressives desperately need to update their understanding of what AI does. 引用 @Noahpinion：Paul Krugman says of AI: "This is a technology of, by and for o
Some telecom network problems don’t have a fix in the runbook. AI agents can investigate, compare fixes and test them in simulation. NVIDIA Nemotron helps agents reason through problems, while NVIDIA OpenShell controls what they can access and do. See how: https://t.co/19eWA7PQZ2
In our 2026 telecom AI survey: • 89% of respondents said open source models and software are important to their AI strategy. • 66% said their company is actively using AI, up from 49% last year. Read the report: https://t.co/qxRZ6dxG1O https://t.co/ak66W8E4Xs
Telcos need AI agents that understand their networks. We provide the full stack to turn open models into specialized agents while keeping operators in control. @ATT, @SoftBank and @Indosat are adopting open models for use cases like autonomous network operations and adapting https://t.co/1fxnyZaVA
Grok 4.7 is excellent at dealing with large code repositories 引用 @XFreeze：Grok 4.7 takes the TOP 3 spots on VulcanBench Frontier v4, outperforming Fable 5.1, Opus 5.5, GPT-6 Astra, GPT-6.1 Sol and other leading frontier models • Grok 4.7 xHigh - 93.15 • Grok 4.7 High - 92.71 • Grok 4.7 Medium - 92
Evaluating Open Models with Artificial Analysis | Nemotron Labs https://t.co/PnydgvUsKM
EmbeddingGemma 2 是一款原生多模态开放模型，面向设备端嵌入。它将代码、图像、音频和视频扩展到文本之外，并统一到共享空间中。
IoT helps people learn. Industrial AI helps systems learn. Physical AI helps the process learn. Azure brings the loop together, and Microsoft has been named a Leader in the 2026 Gartner® Magic Quadrant™ for Global Industrial AIoT Platforms. https://t.co/1e0QHyd5F7 https://t.co/NtqjgDszA0
Google与Constellation公布两项协议细节，合计签约3.6 GW。第一项为890 MW新增核电容量，签订20年购电协议，并涉及伊利诺伊州、宾夕法尼亚州和新泽西州六处站点的11台机组增容，首项增容预计于2028年完成；第二项涉及2,700 MW，原文未提供协议细节。
同一新闻，精选展示《谷歌与Constellation Energy签署20年电力购买协议，拟为PJM新增890兆瓦核电装机》
Kinda interesting that everyone focused on the rolling 12W changes in open vs. closed and when it got most pronounced everyone was freaking out. last two weeks with really good model releases (Opus 5.5, Sol 6.1) closed actually TOOK back share on volume. It's 52% open - 48% https://t.co/cDMNZtwUcd
When The Ads Come to the Data $META $GOOGL full piece: https://t.co/fLIa8v2Gu7 https://t.co/1ItQagXFeg
I had no idea how brave everyone would get just a few years later. 引用 @ByrneHobart：Who will be brave enough to post the first GPT-3 edition of their blog/column/newsletter? Coolest version of this would be a piece in a prestige publication about how AI will never replace writers.
People talk a lot about how the models are extraordinary in verifiable domains (math, coding etc), while not getting much better at things like writing. But is that actually true? Don’t AI-generated stories keep winning prizes and publication in prestigious outlets? 引用 @generativist：the models are
This is from the SA Tokenomics model which is the only granular split I've seen but again this is the 30-40% inference I cite in terms of monetized compute one could make the argument post-training/RL is inference too but if still used for training/internal purposes, I think it 引用 @socialcapital：~
Frontier model subscriptions are heavily subsidised. I made a tool to see how much my API cost would have been. It cost me $35 (and counting) to build this tool. I'm on the $100/mo business premium plan. So far I've spent $4853 in code, not counting chats. It's been three https://t.co/XNldSvjdIN
300 sensors per car. 24 races. Zero margin for error. AWS & @f1 built an agentic AI solution that resolves critical issues up to 86% faster, so engineers focus on the future, not the logs. https://t.co/KNvREXn1wH
Research Spotlight: AI Utilization and Changes in Economic Performance https://t.co/QUAp77mpuL #SCB https://t.co/XeCJUxUB5q
Reason people aren’t controlling the full stack is the capital intensity Hypers will and can underwire the power, shells, chips, etc Others like neos and managed inference provides can barely execute the IT before trying to move upstream This is a capital problem vs lack of 引用 @kellyjgreer：lot
I fully agree here but I think one thing that’s under appreciated is people talking about the duration mismatch Lot of forward credit for TAM penetration x steady state economics when in reality extremely early in usage and adoption That means compute demand going up but it 引用 @nicbstme：This is m
Brother, what will your ARR do That answers your capex question 引用 @firesidealpha：Anthropic's Sholto Douglas says AI capex could rise from $1 trillion this year to $4 trillion in 2028 and begin doubling humanity’s GDP by the early 2030s “Over the course of the last four or five years, we’ve been
You literally could open this app and see 100 of these “The market is asleep at the wheel. I’ve found the next HUGE AI bottleneck. Follow me for more posts like this” *stock up 5,000% in the last 6 months already* **half the time it was literally just pitching nvidia again** 引用 @ParadisLabs：I'm p
I very very very strongly disagree and think that AI adoption in the enterprise beyond coding is basically at the starting line 引用 @krishnanrohit：Everyone uses AI by now. Senior leaders are vibecoding apps. They've adopted claude. They're finding workflows to automate. Folks have their own claws. A
So the AK capital growth loop is the thing that makes this work. However it ignores physics limits on energy and materials and Baumols Cost disease. But perhaps most importantly it ignores the sources and constraints on capital due to cash flow timing deficits Minsky squeeze 引用 @firesidealpha：
A speech model can support Arabic and still struggle with local dialects. Fine-tuning NVIDIA Nemotron 3.5 ASR cut word error rate on Najdi and Hijazi Saudi Arabic from 55% to 30%. Check out our new tutorial and learn how to adapt it for other dialects and languages: https://t.co/a7YcQ08YEs
6️⃣ Non-Associative Algebra: The team worked with Muse Spark to find an exception to a proposed rule about mathematical structures inspired by biology. They went further by developing an alternative characterization, which the researchers checked and refined. Read the paper:
Following gold-medal-level performance from our AI models across five competitions in mathematics, physics, and chemistry, we asked a harder question: can AI contribute when a problem is genuinely open and without an existing solution path? Over the past several months,
Huge thank you to everyone who downloaded Nemotron 3 Diarization and helped it trend on @huggingface! And we appreciate all the comments. @sabbassi_11 answered a few of your questions: https://t.co/7T2tDwhupl 引用 @NVIDIAAI：When several people talk at once, a transcript can get messy fast. Our new