@elonmusk Congratulations on this and all of your Audacious Dreams, Elon - fucking inspirationally incredible! Can't wait to see Boring bore tunnels on the moon, and unique ship designs being launched into space from the Mass Driver.
@GerrardJud46398 @elonmusk 太棒了，你成功着陆了！10. 10. 10。
@elonmusk Awesome and you stuck the landing! 10. 10. 10.
Deployment of 20 @Starlink V3 satellites complete. Today's test will provide critical data as we prepare to expand our Starlink constellation https://t.co/jWKLwXGlsk
❤ 6.7k · 🔁 749 · 💬 771 · 👁 86.5w
热门回复 4
@MarthaJohn1679 @elonmusk 卫星工作了！部署是完美的。
@elonmusk The satellites worked! The deployment was flawless.
@StarSun2 @elonmusk Congratulations !💫🛰️🌸
@theducessR @elonmusk 更多卫星，更多带宽，更好的未来。不断突破界限！⚡🪐
@elonmusk More satellites, more bandwidth, better future. Keep pushing boundaries! ⚡🪐
@elonmusk I hate that you guys don’t show the model iteration in the app. My grok app has been telling me it’s on 4.5 for the past two weeks lmao. You’re the only frontier LLM that doesn’t tell us which version we’re using in the app
Why are we rushing to build multi-billion dollar data centers when the efficiency of local AI is skyrocketing?A high-end Mac with unified memory running a quantized LLM + local RAG solves the data privacy issue entirely, eliminates latency, and provides a massive context window for personal data. Best of all? Zero infrastructure cost to the AI provider.The future of consumer AI isn’t 100% in the cloud—it’s decentralized, localized, and running on user hardware.
Opus got all three right unlike the other models. Houses fly up the funnel and out the top, the wall breaks where the ball hits and the rubble piles up, the bridge drops the truck into the river. Fable had almost nothing on the ground for the tornado to pick up, its building collapsed on its own before the ball even touched it and its bridge blew into sticks all at once. GPT is the cheapest here but its ball never reached the building at all and its bridge fell apart in a way nothing falls apart in real life. Kimi K3, the new Chinese frontier model, ended up in the same place as GPT
@atomic_chat_hq Opus is making me feel fable like an old technology 😭
@Krishna_1896 @atomic_chat_hq 上个月 Fable 5 似乎要毁灭人类，充满戏剧性和禁令，但现在它只是另一个 AI 模型，输给了 K3 和 Opus 5。这真是让人困惑
@atomic_chat_hq Last month Fable 5 was about to destroy humanity, so much drama, bans, now it's just another AI model which is losing to K3 and Opus 5. This is so confusing
@chermdev @atomic_chat_hq 我注意到几点：在 $0.31 的价格下，GPT-5.6 允许更多次迭代，而 Opus 5 一次运行需要 $1.50。此外，20,100 个 token 的价格是 $0.31，意味着 $15/M 个 token，所以这是 Terra 定价，而不是 Sol，考虑到 Fable 5 和 Kimi K3 也被包含在内
@atomic_chat_hq A few points I noticed: At $0.31, GPT-5.6 allows more iteration for the price of one Opus 5 run ($1.50). Also, 20,100 tokens for $0.31 implies $15/M tokens, so that’s Terra, not Sol, considering that Fable 5 and Kimi K3 were included.
@atomic_chat_hq At what point, let's say we fast forward two years, the 50 cents model will be available for 10 cents, and that's how everything gets commoditized. Super excited about that
At Ramp, we tested models on 150k bills submitted by actual businesses, scoring them on whether they predicted every correction a human would make
Grok achieved the highest perfect-extraction rate, beating similarly priced models Gemini Flash 3.6, GPT 5.6 Terra, and Sonnet 5.
This is a demanding long-context reasoning task. The model must infer patterns across 100K+ tokens of prior invoices, business memories, and human corrections, then apply them to new bills. The goal: zero-click accounts payable, with invoices processed correctly without human intervention.
❤ 1.7k · 🔁 265 · 💬 469 · 👁 70.6w
热门回复 4
@AlexeyShumarin @elonmusk @grok 兄弟：
AI 掌控地球的概率有多大，当前世界上正在发生的所有战争都是由它引起的呢？
@elonmusk @grok bro:
What is the probability that AI has seized power on planet Earth and all the wars in our world that are happening right now are its merit?
@TomGiant1 @elonmusk 我的日常使用的 Hermes Agent 🔥🔥🔥🔥
@elonmusk And my daily driver in Hermes Agent 🔥🔥🔥🔥
@MendhamMike @elonmusk 开放模型将权力交给用户。在封闭模型中，一个秘密团队根据他们自己的偏见决定什么是正常的，什么需要微妙的重构。这可能在人口层面上导致隐秘、愤世嫉俗的滥用。只需问问 ChatGPT 关于功绩和 DEI 或犯罪统计数据。https://t.co/sUQ9Z0JB3P
@elonmusk Open models give the power to the user. In closed models a secret team with their own priors decides what is normal and what requires subtle reframing. That has the potential for insidious, cynical abuse at a population level. Just ask ChatGPT about merit and DEI or crime stats. https://t.co/sUQ9Z0JB3P
@DrKnowItAll16 @elonmusk Yes indeed!
@jtalbott @elonmusk 👏 现在你们需要抓住Spark DGX开始行动了。
@elonmusk 👏 now yall need to grab a Spark DGX and get going.
@iScienceLuvr To a layman, this means that AI is not a "product" to be trademarked or copyrighted, but a technology like the printing press to be shared or improved, then used to produce other profit-making (?) products. Scant consolation for the companies that invented the technology. 1/
2. AMD $AMD CEO Lisa Su 在今天的 AMD Advancing AI 活动上表示，AI 加速器市场到2030年可能增长至1.4万亿美元，而在更广泛的计算市场中，预计将达到2万亿美元。她还指出，过去两年中 AI token 消耗量激增了158倍，强烈凸显计算需求的快速扩大。AMD 推出了 Helios，这是一种全新机架级 AI 系统，围绕 MI450 加速器构建，Su 称其为行业内最快的 AI 加速器。Helios 已全面投入生产，预计第三季度底开始发货，并在第四季度加速。
12. Meta 的 $META 新的120亿美元得克萨斯数据中心融资计划，与近1GW 的项目相关，据报道正在以7% 以上的收益率进行讨论。这大约是比 Meta 9个月前的270亿美元 Hyperion 融资高出40个基点，增加了约4800万美元的年度利息支出。Hyperion 债券现在交易在每美元0.96元左右，显示出大型 AI 基础设施项目的融资成本正在开始上升。
华尔街是地球上最精彩的表演。
展开原文
A TON OF THINGS HAPPENED IN THE STOCK MARKET TODAY.
Here's a full recap:
1. Tesla $TSLA was down 14% today, its worst day since January 2024, while Google $GOOGL was down 7% after posting its first negative free cash flow quarter since 2004. The macro backdrop is not helping either. Crude oil has pushed back above $90/barrel for the first time since June and is up 25% over the past month. At the same time, the 10-year Treasury yield has crossed 4.7%, while jobless claims came in at 187K vs 211K expected, showing a stronger labor market but also raising concerns that inflation pressure could persist. Today was the worst day for the Mag 7 since April 2025 with all 7 stocks erasing $800B of market cap.
2. AMD $AMD CEO Lisa Su said at the AMD Advancing AI event today that the AI accelerator market could grow to $1.4T by 2030, within a broader compute market expected to reach $2T. She also noted that monthly AI token consumption has surged 158x over the past two years, underscoring how quickly compute demand is scaling. AMD unveiled Helios, its new rack-scale AI system built around the MI450 accelerator, which Su called the fastest AI accelerator in the industry. Helios is now in full production, with shipments expected to start at the end of Q3 and ramp through Q4.
3. Intel $INTC posted a strong Q2, with revenue of $16.1B vs $14.5B expected, up 25% YoY, and adjusted EPS of $0.42 vs $0.22 expected. Adjusted gross margin came in at 41.8% vs 39% expected, up 1,210 bps YoY, while non-GAAP operating margin reached 17.2% vs 10.7% expected. Segment results were strong across the core business, with CCPG revenue up 13% YoY to $8.9B, DCAI revenue up 59% YoY to $6.3B, total Intel Products revenue up 28% YoY to $15.1B, and Intel Foundry revenue up 31% YoY to $5.8B. Management said Q2 marked Intel’s strongest revenue growth in more than 15 years, driven by better execution, higher factory yields, improved cycle times, and stronger customer demand.
4. Trump warned that the U.S. will hold Iran responsible if the Houthis resume attacks on commercial shipping. He said the Houthis had acted “responsibly” after previous U.S. strikes, but are now “starting up again” after reportedly firing on two Saudi Arabian ships last night. Trump said the Houthis are a proxy of Iran, and that any further attacks would trigger major military punishment against both Iran and the Houthis.
5. Nvidia $NVDA is committing $1.5B to Amkor $AMKR through a prepayment tied to a multi-year advanced packaging and development agreement. The funding will help Amkor expand U.S. packaging capacity at its Arizona campus as both companies work on next-generation packaging and test technologies for AI and accelerated computing. The partnership will focus on high-density interconnects and heterogeneous integration, which are key for combining multiple chips and components into more powerful systems.
6. The top 10 most active options today by contracts traded were $TSLA with 3.6M contracts, $NVDA with 2.4M contracts, $GOOGL with 994K contracts, $AMZN with 940K contracts, $SPCX with 917K contracts, $AAPL with 732K contracts, $INTC with 706K contracts, $MU with 664K contracts, $GOOG with 560K contracts, and $MSFT with 532K contracts.
7. South Korea is moving to tighten rules around leveraged single-stock ETFs and ETNs beginning July 31. Retail investors will now need about $20,300 in cash to open or add to a position, up from roughly $6,800 previously. The key change is that stocks, bonds, and other securities will no longer count toward the deposit requirement, making it harder for retail traders to access these leveraged products.
8. U.S. mortgage rates rose for the third straight week, with the average 30-year fixed climbing to 6.58%, putting rates back near their highest level in a year.
9. AMD $AMD and Cerebras $CBRS are teaming up on a disaggregated AI inference architecture that divides workloads between both platforms. In the setup, AMD Helios manages prompts and long-context processing, while Cerebras’ Wafer-Scale Engine focuses on ultra-low-latency token generation. The companies say the combined system can deliver up to 5x more tokens per second per watt than Cerebras alone, with initial availability expected through Cerebras Cloud in the second half of 2026.
10. Uber $UBER cut 10% of roles in its customer service operations as it restructures the division and leans further into AI. The company said fragmented workflows were making it harder to roll out AI at scale, and this marks Uber’s first layoff round specifically tied to AI-driven efficiency.
11. SpaceX $SPCX is reportedly turning away satellite operators looking for dedicated Falcon 9 launches beyond 2028 as the company shifts more of its long-term launch strategy toward Starship, per Bloomberg. SpaceX has also stopped taking future Falcon 9 rideshare reservations and has paused production of some expendable Falcon components, though Falcon 9 is still expected to remain active for NASA and Pentagon missions. The risk is timing: if Starship is not commercially ready by 2028, the market could face a major launch-capacity gap, creating a bigger opening for competitors like Rocket Lab, Blue Origin, and ULA.
12. Meta’s $META new $12B Texas data center financing, tied to a nearly 1GW project, is reportedly being discussed at yields above 7%. That is roughly 40 bps higher than Meta’s $27B Hyperion financing from just nine months ago, adding about $48M in annual interest expense. Hyperion bonds are now trading around 96 cents on the dollar, showing how financing costs for massive AI infrastructure projects are starting to move higher.
@unusual_whales It’s less valuable now because for the first time in company history they’re losing money and they went from $90 billion cash in the bank to $90 billion bank loan.
And Google Gemini still has real problems telling the truth.
Unitree Super Athlete AS2-W 😉 Fluid movement, compact yet exceptional. Supports a continuous 16kg payload, with 30+ km unloaded driving range https://t.co/AwjdwnzUFs
🔴 中国机器人制造商 Unitree 推出能够在崎岖地形中行走、攀登陡坡和承载负载的人工智能四足机器人 Super Athlete AS2-W https://t.co/q5RFcgLSrY
展开原文
🔴 Çinli robot üreticisi Unitree, zorlu arazilerde koşabilen, dik yamaçlara tırmanabilen ve yük taşıyabilen yapay zekâ destekli dört ayaklı robotu Super Athlete AS2-W'yi tanıttı. https://t.co/q5RFcgLSrY
@PiyasaTturkiye Robot efsane de, altta Çinlilerin yorumlara bakın hindistanlılar gibi herkese saldırıyorlar. Türklerle ne dertleri varsa kimse Çinlilere bişey dememiş onlar kendi kendilerine hakaret ediyor.
@BogdanT_R @PiyasaTturkiye 试着逃避那东西 😅
@PiyasaTturkiye Imaging trying to outrun that thing 😅
By the time the deadline hit, the app was already back. On a network with no company to send the order to.
Here's what happened:
• India orders GitHub to remove BitChat, citing "unlawful communication" • GitHub has 3 hours to comply or face legal action • Within hours, a developer mirrors the entire codebase on Radicle • Radicle has no central server. No company. No CEO to subpoena. No app store to pressure. • The instructions go out: clone it, seed it, keep it alive • Every person who seeds it becomes another copy the government can't touch
The part nobody's saying out loud: this is the exact scenario BitChat was built for.
It runs on pure Bluetooth. No internet. No SIM. No phone number. It was already unkillable at the app layer, that's why protesters in Nepal, Iran, and Madagascar used it to route around government shutdowns.
Now the SOURCE CODE is unkillable too.
India didn't delete BitChat. It just handed the entire internet a live demo of why decentralized infrastructure exists.
You can order a company to take something down. You can't order a network that isn't owned by anyone.
Command to mirror it yourself is already public. People are seeding it right now.
@Crypto_Jargon 🚨🇮🇳BREAKING: 印度刚刚在三小时内命令 GitHub 下架 Jack Dorsey 的 BitChat。
中国今年已经两次从其应用商店下架 BitChat，原因是伊朗等地区的抗议活动协调。印度以公共秩序和非法集会风险为由援引 IT 规则。
Jack Dorsey 的说法是：政府不喜欢无法关闭的技术。
你可以关闭移动数据。但无法关闭站在一起的两部手机。
🚨🇮🇳BREAKING: India just ordered GitHub to take down Jack Dorsey's BitChat within three-hours.
The mechanism explains why. BitChat runs entirely over Bluetooth, no internet, no servers, no phone number needed. That's exactly why NEET protesters started using it after mobile internet got suspended around Parliament and Jantar Mantar this week.
China already pulled BitChat from its App Store twice this year for the identical reason, protest coordination during Iran and elsewhere. India's citing public order and unlawful assembly risk under IT Rules.
Jack Dorsey's saud: the government doesn't like technology it can't turn off.
You can shut down mobile data. You can't shut down two phones standing next to each other.
@Crypto_Jargon This is the whole test, right? No server, no company, no one to send the takedown to. That's decentralization — can anyone flip the switch or not. We score crypto on exactly that: $DGB sits at 96, node spread too wide to pull the plug. https://t.co/Jjt4XFZTqK
几年前，这意味着需要花费数周进行建模、脚本编写、照明和动画。现在你可以描述世界，看着 Kimi 构建它，然后花时间修复最终10% 而不是手动构建前90%。
Kimi K3 + Blender MCP 基本上是文本到3D 的转换，而不会将结果困在无用的生成视频中。每个物体、材质、灯光、摄像机和关键帧都是可编辑的。
展开原文
Blender just became a prompt box.
With Kimi K3 connected through Blender MCP, you can describe a scene like this in plain English and let the model handle the ugly part: terrain, buildings, lighting, materials, camera movement, animation, and the Python scripts holding everything together.
The interesting part is not the first render. Kimi can inspect the result, notice that the camera clips through a tree or the city looks suspiciously like plastic, then edit the actual Blender scene and render it again instead of restarting from zero.
You still need taste, because “make it cinematic” remains one of humanity’s least useful instructions. But the distance between an empty Blender file and a fully editable 3D world just got embarrassingly small.
A few years ago, creating this meant weeks of modeling, scripting, lighting, and animation. Now you can describe the world, watch Kimi build it, and spend your time fixing the final 10% instead of manually constructing the first 90%.
Kimi K3 + Blender MCP is basically text-to-3D without trapping the result inside a useless generated video. Every object, material, light, camera, and keyframe stays editable.
This scene was created by Dylan Donaldson (@Zertox3D), a Generalist Artist at ILM.
He built it using Blender and other software and rendered it with Karma in Houdini. He's also posted a full breakdown of the shot, so it's clear this came from his own creativity and effort.
@DilumSanjaya with the claude and other model i tried to build similar like it for training the UDRF model in issac sim that literal failed and can't handle that complexity but this is insane ,and i love the taste in design
@Gowtharr @DilumSanjaya 希望能看到有人创造新的东西。每个当前 LLM 都是基于训练数据中过度泛滥的内容或网络/书籍中现成的资料。想象一个 n-pod 来浏览印度某个城镇的 3D 扫描，让模拟输出各种变化及其优化。
@DilumSanjaya Would love to see someone make something new. Every cad llm is of something overpopulated in the training data or readily available on the webs/books. Vibe a n-pod to walk through a 3d scan of a town in India, have the sim output the variations and their optimizations.
SK Group 和 NVIDIA 今天宣布了一项超过5000亿美元的计划，涵盖 AI 工厂和下一代存储技术。
➡️ @SKtelecom 正在韩国建设一座2吉瓦的 NVIDIA Vera Rubin DSX AI 工厂，以满足全球计算需求。 ➡️ @SKhynix 和 NVIDIA 将共同开发下一代 AI 存储技术，包括 HBM。
韩国已准备好通过拥有世界一流的网络和数据中心、芯片技术领导地位以及庞大的产业规模，成为全球 AI 强国。
展开原文
SK Group and NVIDIA today announced a $500 billion-plus initiative spanning AI factories and next-generation memory.
➡️ @SKtelecom is building a 2-gigawatt NVIDIA Vera Rubin DSX AI Factory in Korea to serve global compute demand. ➡️ @SKhynix and NVIDIA will codevelop next-generation AI memory, including HBM.
Korea is ready to become a global AI powerhouse with world-class networks and data centers, leadership in chip technology and vast industrial scale.
@nvidia @SKtelecom @SKhynix @JensenHuang hope your major partner @Supermicro $SMCI will be involved to build and deliver the rack servers for this AI factory / data centre. $SMCI.
$NVDA AND SK GROUP UNVEIL AI INITIATIVE VALUED AT OVER $500B
该协议使 Nvidia 能够获得 SK Hynix $SKHY 存储技术，而 SK Telecom 计划建设一座2吉瓦的 AI 数据中心，由 Nvidia 的 Vera Rubin 芯片和 SK Hynix HBM4 提供支持 https://t.co/gOXm868bI9
展开原文
$NVDA AND SK GROUP UNVEIL AI INITIATIVE VALUED AT OVER $500B
The agreement gives Nvidia access to SK Hynix $SKHY memory, while SK Telecom plans to build a 2-gigawatt AI data center powered by Nvidia’s Vera Rubin chips and SK Hynix HBM4. https://t.co/gOXm868bI9
❤ 910 · 🔁 104 · 💬 23 · 👁 12.2w
热门回复 4
@dolores_capital @wallstengine $SKHY会在周一再下跌7%对吧
@wallstengine $SKHY going to be down another 7% on monday with this huh.
