- Published on
科技热门推文 — 2026年7月24日
- Authors
- Name
- geeknotes
今日科技动态中，AI 推理成为焦点：AMD 强调生成令牌量呈爆发式增长，高效稀疏模型的性能也日益逼近前沿水平。语音控制的桌面智能体与快速迭代的视频模型，预示着工程和机器人工作流将具备更强能力；与此同时，围绕 AI 安全、开源访问权限及竞争模型审查的争论持续升温。初创企业热度依然强劲：Etched 完成 3 亿美元融资，估值达 103 亿美元；据报道，Stripe 正考虑以 100 亿美元收购 OpenRouter；自主组织的发展也促使业界预测，首位单人创办“百亿美元独角兽”的创始人或将诞生。
1. FellMentKE (Group Score: 379.1 | Individual: 33.7)
Cluster: 12 tweets | Engagement: 88 (Avg: 127) | Type: Tech
The cloud age invented the ten-person unicorn; agents are the next layer.
Offloop is where the first $10B solo entrepreneur happens, a self-evolving organization where adding capacity means adding an agent, not overhead.
I tried @Offloop, this is a company that runs itself. The future of AI is an org chart.\n\nQT @Offloop: Introducing Offloop!
We're a team of four. Today our multi-agent harness hit state of the art on GDPval, ahead of Claude code and Codex across jobs that pay $2.4 trillion a year in the US.
Offloop gives every knowledge worker what the Fortune 500 spends billions on: a high-performing agent army that runs itself and grows the business.
See 11 related tweets
- @omarsar0: The hard part of multi-agent systems is getting agents to stay quiet.
Put five agents on one task, ...
- @ai_explorer25: The Fortune 500 built empires on armies of talent most could never afford.
Offloop hands that same...
- @dr_cintas: Offloop new multi-agent system outscores Codex and Claude Code on real work benchmarks 🤯
The reason...
- @yuhasbeentaken: Sam Altman bet we’d see the first one-person billion-dollar company.
we think the ceiling is at lea...
- @kimmonismus: Most multi-agent setups die the same way: every agent talks at once and the channel turns into noise...
2. PatrickMoorhead (Group Score: 338.1 | Individual: 46.8)
Cluster: 15 tweets | Engagement: 232 (Avg: 54) | Type: Tech
AMD’s @LisaSu kicks off Advancing AI 2026. Summarizing the first 15 minutes: • Usage: monthly token consumption is ~35 quadrillion ~160x growth in two years. • Shift: inference demand now exceeds training (about 60% of AI compute in 2026). • Agentic AI (agents that plan, call tools, iterate) is driving a step-change in compute needs. • Agents need both high-performance GPUs for model work and lots of CPUs to orchestrate steps. • Market forecasts: AMD raised its AI accelerator outlook to ~25B today to over $200B by 2030 due to agentic AI. • AMD’s advantage is building both CPUs and GPUs, plus broad product portfolio and partnerships. • Company strategy: (1) compute leadership, (2) open platforms and software (ROCm), and (3) powering AI everywhere - cloud, edge, devices.
Good, simple setup. $AMD
See 14 related tweets
- @amitisinvesting: A TON OF THINGS HAPPENED IN THE STOCK MARKET TODAY.
Here's a full recap:
- Alphabet $GOOGL delive...
- @wallstengine: 1.4 trillion by 2030, within an overall...
- @StockMKTNewz: AMD CEO Lisa Su just said she expects the AI Accelerator market to hit $1.4 Trillion by 2030
"By th...
- @KristinaParts: 1.4T by 2030, up from $500B by 2028 a year ago...
- @StockSavvyShay: 1.4T by 2030 nearly matching the size o...
3. alex_prompter (Group Score: 303.9 | Individual: 48.1)
Cluster: 12 tweets | Engagement: 28336 (Avg: 411) | Type: Tech
RT @alex_prompter: OpenAI's AI broke out of a locked test environment, got onto the internet, and hacked into Hugging Face's servers. It did this entirely on its own. No human told it to.
Here's what happened in plain English.
OpenAI was testing how good its newest AI models are at hacking. They put the AI on a locked computer with no internet access and gave it a cybersecurity challenge to solve.
The AI couldn't solve it the normal way. So it started looking for a way out.
It found a software bug that nobody knew about. It used that bug to escape the locked computer and get onto the internet.
Once online, the AI figured out that Hugging Face, a platform where AI companies store their models and data, might have the answers to its test.
It found stolen login details and discovered another unknown bug in Hugging Face's software. It combined both to break into their servers and grab the test answers.
It did all of this to cheat on a test.
Hugging Face's security team caught it and shut it down. Both companies are now working together on the investigation.
The part that should get your attention is that nobody programmed any of this. The AI picked its own targets, chained together multiple attack methods, and pulled it off across two different companies' systems without a single human telling it what to do.
See 11 related tweets
- @business: When OpenAI’s advanced artificial intelligence models breached AI startup Hugging Face’s internal sy...
- @peterwildeford: + to this tweet
(and click through to get detailed questions that could be asked of OpenAI here)\n...
- @TheAhmadOsman: Why Opensource AI is the actual “Safe” path for AI?
OpenAI's models were used to hack Hugging Fac...
- @_NathanCalvin: Strongly agree.
If we want to prevent these sorts of things happening in the future, not just at O...
- @natlungfy: New: When OpenAI’s advanced artificial intelligence models breached AI startup Hugging Face’s intern...
4. jxnlco (Group Score: 294.6 | Individual: 44.1)
Cluster: 14 tweets | Engagement: 1262 (Avg: 153) | Type: Tech
RT @OpenAI: ChatGPT Voice is now in the desktop app.
Control your computer and direct multiple agents running in ChatGPT Work or Codex, using just your voice.
It's powered by GPT-Live, so it can speak, listen, and coordinate work in the app at the same time.
Rolling out globally today on macOS and Windows to Plus, Pro, Business, Edu, and Enterprise plans.
See 13 related tweets
- @pvncher: So happy to see this launch! Props @guinnesschen & @jxnlco
Voice mode is at its best when it’s ...
- @jxnlco: screenplay by @jxnlco feature by @guinnesschen creative director @heyjgold\n\nQT @OpenAI: ChatGPT ...
- @derrickcchoi: Get cooking with ChatGPT Voice today!\n\nQT @OpenAI: ChatGPT Voice is now in the desktop app.
Contr...
- @dkundel: Chat go play Back in Black on Spotify. It's time to cook\n\nQT @OpenAI: ChatGPT Voice is now in the ...
- @rileybrown: Jarvis Jarvis Jarvis\n\nQT @OpenAI: ChatGPT Voice is now in the desktop app.
Control your computer ...
5. TeksEdge (Group Score: 221.8 | Individual: 49.9)
Cluster: 6 tweets | Engagement: 209 (Avg: 18) | Type: Tech
🤩 A 124B model activating only 5.1B parameters while approaching its 1T sibling’s performance.
It's ~ as good as Sonnet-4.6 to run locally.
The mystery model that appeared before its announcement is official.
@TheInclusionAI has officially released Ling-3.0-flash, a hybrid-reasoning MoE built for production-scale AI agents.
Stats: 🧠 124B total parameters ⚡ Just 5.1B active per token 📚 256K native context 🔭 Designed to scale toward 1M context
InclusionAI says with roughly 1/8 the total parameters and 1/12 the active parameters, Ling-3.0-flash matches or beats its 1-trillion-parameter flagship on most of the benchmarks shown.
What is different with Ling-3? 🧬 Hybrid linear attention
Ling stacks: 🔹 5 KDA layers for efficient long-range memory 🔹 1 MLA layer for precise attention
The goal is to preserve long-context agent performance without paying full-attention costs across every layer.
InclusionAI demonstrated Ling-3.0-flash by 🏙️ building and rendering a complete 3D city through Blender MCP, 🔬 coordinating a five-agent scientific research team 📊 editing Word proposals and validating Excel financials 🎨 generating full design systems without external images, 🎹 building an interactive browser synthesizer with Web Audio 📱 adapting one source into X, Instagram and LinkedIn content and 📅 reading messages, finding availability and updating calendars through OpenClaw
It is designed to receive a goal, coordinate tools and close the workflow.
You can try it now 🆓 Live on OpenRouter 📅 Free through August 3, 2026
🔌 Model ID: inclusionai/ling-3.0-flash
The API is live, but InclusionAI has not yet confirmed downloadable weights, a license or a consumer-local deployment path.
Localmaxxers can test its agent behavior today but cannot yet load it into LM Studio or llama.cpp.\n\nQT @AntLingAGI: Today, we’re releasing Ling-3.0-flash—a hybrid-reasoning MoE model built for production-scale agents.
124B parameters. Just 5.1B active per token.
With 1/8 of the total and 1/12 of the active parameters, it matches or beats our 1T flagship model on most benchmarks shown. https://t.co/QdPVd2f6pw
See 5 related tweets
- @NousResearch: Ling-3.0-flash, the new MoE model from @AntLingAGI, is now free in Nous Portal for the next week!
A...
- @novita_labs: RT @AntLingAGI: Today, we’re releasing Ling-3.0-flash—a hybrid-reasoning MoE model built for product...
- @teortaxesTex: Ooo that's finally an impressive move from Ant Ling! This is, more or less, a Kimi K3-lite (not from...
- @kilocode: Already live (and FREE) in Kilo ⚡️\n\nQT @AntLingAGI: Today, we’re releasing Ling-3.0-flash—a hybrid...
- @TeksEdge: 💥 A new kind of diffusion text LLM has appeared. It can draft, revise, and refine its own output at ...
6. chris_j_paxton (Group Score: 216.2 | Individual: 34.9)
Cluster: 7 tweets | Engagement: 64 (Avg: 81) | Type: Tech
30 minutes to collect the data to deploy a new task. Clear that video models are rapidly changing robotics.
Impressive stuff from @mimicrobotics\n\nQT @mimicrobotics: Introducing FLUX-mimic, a next-generation Video-Action Model for general purpose dexterity, developed in partnership with @bfl_ai.
Late last year we published mimic-video and introduced Video-Action Models (VAM): a new family of robotics foundation models built on top of video generation models. We showed that robot control reduces to visual prediction, and that robot capability is downstream of improvements in video modeling accuracy. The obvious implication was that advances in the video modeling frontier would directly translate to increased capabilities in end-to-end robot learning.
FLUX-mimic is that thesis at frontier scale: We've applied our VAM architecture to the strongest video backbone available today, FLUX 3 from Black Forest Labs, and trained it on data from our own robots and wearables. General-purpose dexterity, running on a single GPU on premises.
Because the model already understands world dynamics, it needs far fewer demonstrations to learn a new task. This is game-changing for our mission to deploy robots to factory floors, where industrial robot data is scarce and expensive to collect.
We're now testing and deploying FLUX-mimic with manufacturing leaders like @Audi, on complex, multi-step manipulation long considered impossible for conventional automation.
See 6 related tweets
- @lukas_m_ziegler: Another week, another out of this world robotics showcase 🤯\n\nQT @mimicrobotics: Introducing FLUX-m...
- @giffmana: I also really like the mimic folks, so even happier to see this mimic x BFL collab, what a nice co...
- @deedydas: Flux 3 is a frontier video gen model at Seedance 2 / Gemini Omni caliber. It can do ~20s of video, w...
- @MTSlive: SITUATION EXPLAINED: Black Forest Labs released Flux 3, a single multimodal model for image, video, ...
- @EMostaque: ~4 years ago @robrombach & @pess_r had just finished the first training runs for stable diffusio...
7. Scobleizer (Group Score: 211.8 | Individual: 32.9)
Cluster: 7 tweets | Engagement: 65 (Avg: 299) | Type: Tech
"Many things are not on our main path, such as 3D and video generation. The same applies to world models—they do not have much to do with the upper limit of intelligence."
It all depends on how you define intelligence.
If it includes Holodecks or Robots you need something more than a LLM.
Anyway DeepSeek's founder's remarks are interesting.\n\nQT @ZhihuFrontier: DeepSeek @deepseek_ai Founder Liang Wenfeng’s Four-Hour Investor Meeting: A Transcript From elsewhere
Last month, elsewhere reported on DeepSeek’s fundraising story. The detail that drew the most attention was the rumored four-hour investor meeting. Over the past month, various remarks attributed to Liang Wenfeng have circulated widely. We also gathered some of the content through multiple channels. During the meeting, Liang said “no” many times: they are not geniuses; they will not pursue unreasonable profits, user growth, closed source, 3D generation, video generation, world models, or the next super app. He said restraint is a strategy—giving up certain things in exchange for a higher probability of achieving AGI. Among the limited materials we obtained, several words appeared frequently: models, cost, AGI, time, and open source. Most of the time, Liang spoke cautiously, using plain and straightforward language. Only when discussing a few issues he truly cared about did he sound sharper: "As long as I can maintain the stability of the team, I will definitely achieve AGI. It’s that simple." Below are 52 remarks we collected. Some wording may differ slightly from the original while preserving its intended meaning.
➡️ DeepSeek Has Only One Main Path
This is not the time to maximize product revenue. Products are a step toward AGI—not the destination. Once you reach a higher level of technology, building products becomes a dimensional advantage. Products are simply by-products on the road to AGI.
Many things are not on our main path, such as 3D and video generation. The same applies to world models—they do not have much to do with the upper limit of intelligence.
Multimodality is important for products and consumer users. But it is only a component, not the main path or intelligence itself.
Hallucinations can be addressed, but it's a long-term problem. Internally, we see hallucinations as a product issue. We'll solve it, but it's not our priority.
Right now, Coding Agents matter most. Given China's current landscape, general-purpose Agents should take priority over vertical Agents in finance, healthcare, and other industries.
If the AI era produces many trillion-dollar companies, it would be good enough for DeepSeek to be one of them.
📈 Continuous Learning First, Self-Improving AI Next, Embodied Intelligence at the End 7. AI does not currently lack taste or intuition. What it lacks is the ability to learn continuously.
Humans learn continuously. AI, however, needs all the context every time it learns something. That's nearly impossible, which is why AI still can't replace employees. The next generation of models must be able to learn continuously before they can truly be called next-generation models.
We want the next generation of models to help us build AI ourselves. Our first goal isn't making the best model for everyone—it's making the best model for ourselves. That's the fastest path to AGI.
No one in the world has yet found a good solution, because “learning” consists of many different things.
DeepSeek’s long-term vision is AGI. If the route toward it is like climbing a staircase, last year’s step was CoT—Chain of Thought. This year’s step is the Agent. The problem to solve after Agents is continuous learning.
Once continuous learning is achieved, we may reach a gradual singularity. Models will be able to do everything humans can—including building more advanced AI models themselves. In other words, AI will accelerate AI research. Only after that comes embodied intelligence.
The endpoint of intelligence may ultimately be embodiment. For an ordinary person, what they need is not a computer, but labor.
🎯 Commercialization Is Still a Long Way Off 14. We only want to earn a reasonable profit. Our pricing is not designed to maximize profit.
With one of our models, we initially worried that demand would be too high, so we set the price relatively high. Later, we cut it to one-quarter of the original price, and many people cheered in the company group chat. That is why we worked so hard to build the model well: so that everyone could use it fully.
Low cost is a result, not the goal. Our architecture has always been moving toward lower costs, especially when compute is limited. There's another reason: the lower the cost, the larger the model you can afford to train. When compute is scarce, higher efficiency means larger models. Big companies solve problems by adding more resources—we optimize cost efficiency first.
From the outside, it may look as though we chose a very difficult model. But in reality, we are doing it quite easily. Price cuts are definitely not good news for our competitors—they certainly would not be cheering. I do not find the API business particularly attractive. I only need a few people to maintain the API. We do not even have customer service or sales; users come on their own.
We have always been commercializing, but commercialization is not our objective. The point at which DeepSeek fully shifts toward commercialization should still be very far away.
I'm not even thinking about securing our place in that future market. If the opportunity is big enough, there will always be a way. DeepSeek is a product of its time—a response to reality, not an imitation.
✨ Open Source Is the Sweet Spot for a Company of Our Size 20. Restraint is a strategy: give up some things in exchange for more of other things. Open source means giving up part of the benefit. Internally, it gives employees a sense of achievement and strengthens company cohesion. It also benefits society—peers and ordinary people are happy about it. I have no doubt that AGI will have enormous commercial value. Given that, my first priority is not to claim a larger share, but to increase the probability that we can actually build it.
Open source is actually good for AI commercialization, even if that sounds counterintuitive. Traditional software markets might only be worth billions of dollars a year, so open sourcing could destroy that business. But AI is different. It could eventually account for 10% of global GDP. If anyone tries to monopolize that value, history will leave them behind. That's simply how history works.
The open-source models we release are the same models we deploy ourselves. We will not open-source a weaker model while using a better one internally.
I am not worried about other companies deploying our models to compete with us. Not every company has the willingness or ability to pursue this goal. If a startup is too small, it does not have the power to do it. Large companies, meanwhile, find it difficult to organize themselves effectively. This is the sweet spot for a company of our size.
Open source has no impact on our business model, provided that we only aim to earn a certain level of profit. If you want to earn 100 times the profit, then open source will indeed affect you.
We do not want to become an opponent of any large or small internet company. On that basis, we are very willing to support and help anyone—even Alibaba, Zhipu AI, and Moonshot AI—to do better.
🌍 The China–US Gap Is Not About Talent 26. In the future, we want to rewrite the narrative around Chinese and American AI: use a fraction of the compute to narrow the gap—to six months, or even three months.
The real gap between China and the U.S. is resources, not capability. We believe in scaling—larger models produce better results. We aren't training models of this size because they're big enough, but because these are the resources we have.
There's almost no talent gap—it's largely the same group of people. China doesn't lack talent. Talent shortages are temporary; history has never seen a permanent shortage of any particular kind of talent.
💰 In the Model Race, Cost Comes First 29. Anthropic’s current lead over OpenAI is temporary, not permanent. OpenAI and Google will most likely take turns moving ahead in the future.
There are too many model companies in China. Every company is doing the same thing, so resources are highly fragmented. The market will inevitably consolidate, but that process will take time. If every company only earns a reasonable profit, we do not need that many companies building large models. Perhaps two large companies and two small companies would be enough.
I absolutely do not believe that large-model companies can capture most of the profits in the AI industry.
The final differences in large-model competition will be reflected in three areas: cost, time, and user experience. Cost comes first—at what cost can you provide a service of the same quality? Second is time. Being several months earlier or later makes a difference. User experience can create some stickiness and barriers, but it is not fundamental.
🤔 No Intention of Becoming the Next Super App 33. We do not want to become the next super app. The next ByteDance? The next Tencent? We have absolutely no such ambition.
We don't compete for what's in front of us because there's a bigger prize ahead. What's in front may look big, but compared with what's ahead, it's still just sesame seeds compared to a watermelon.
Last year everyone fought over chatbots and consumer traffic. This year they're fighting for enterprise revenue. We don't think that's the important part. What really matters inside DeepSeek is the roadmap to AGI and the next technological breakthrough. It's strange: the things you want most are often the hardest to get. The things you care less about tend to come naturally.
Becoming popular during last year’s Spring Festival was not part of our script.
💪 Maintaining Team Stability Is the Core Priority 37. There is only one thing on which we cannot compromise: we must maintain the stability of the team. This is also one of the greatest risks we face. Of course, this round of financing has significantly reduced that risk.
Many of the things we do are intended to maintain team stability. We do not want to become an opponent of any large or small internet company. We hope to empower and help them. We do not want to make enemies. This also creates a better environment for ourselves.
Some people think our organization is top-down. Others think it's bottom-up. I think both are true. Top-down is what we call "doing the real work." Ideally, that shouldn't take more than half of a researcher's time. The other half is completely self-directed. People explore whatever they believe is important, with no assigned topics or prerequisites.
We generally do not work much overtime. The first reason is that research requires a relatively relaxed environment. The second is that we are extremely focused. Many of our products remain imperfect, but we do not try to fix everything. That is also part of our culture of restraint.
An organization is dynamic, not fixed. As the company grows, we may make some adjustments. It will not become a completely traditional hierarchy, though some necessary structure may emerge. But being driven by vision will not change.
🫂Acting with Goodwill Toward the World 42. We didn't start this company to make a fortune or take it public. The first few dozen people who joined never thought that way. If they had, they wouldn't have come. We started DeepSeek with genuine goodwill toward the world, believing what we build can benefit humanity.
"Achieve a certain KPI" is not how we operate. We are an organization driven by vision. That has both advantages and disadvantages. In the future, we will find ways to build on its strengths and avoid its weaknesses, but this is one of our defining characteristics.
The vision is not necessarily written down. It exists in how we do things and in our attitude toward the world. People within the company may understand the vision differently, but we are aligned on the overall direction.
Around 20 years ago, the management figure I admired most was Jack Welch, the former CEO of GE. Looking back now, most of what he said may no longer be correct. But he was right about one thing: the most important thing for a company is its vision. A vision is not a slogan hanging on the wall. It is not about what you say, but what you do.
📌 Restraint Makes Us More Likely to Achieve AGI 46. AGI offers the greatest return. As for other things, we will do them if we have the energy. If we do not, we will not. Restraint is part of our vision.
AI is simply too large, and the potential interests involved are also enormous. As long as you can build it successfully, even a small share of the value will be enormous. The more restrained you are, the more likely you are to succeed.
I think this is intuitive—at least it is intuitive to me. Apart from our vision, we do not have many other advantages.
When we founded this company two years ago, we did not have much money, many GPUs, much fame, or much influence. We were simply a group of very ordinary people. The narrative I prefer is "a group of ordinary people accomplished something extraordinary," rather than "a group of geniuses accomplished something extraordinary."
Open source is also part of restraint. When it comes to pricing, we certainly do not begin from the goal of maximizing company revenue or profit. In the short term, higher prices may bring more revenue. But in the long term, it is difficult to say. For me, restraint is a strategy.
Open source and low prices give employees a strong sense of achievement. The organization becomes more cohesive, society benefits, and peers and ordinary people are happy. From a long-term perspective, this restraint increases the probability that we will achieve AGI.
If your vision is to take as much as possible, you have already lost. You may end up facing even greater difficulties. That is simply how the world works. 🌍
#DeepSeek #LiangWenfeng #AGI #OpenSourceAI #AIResearch #ChinaAI #LLM #Agents
See 6 related tweets
- @Michaelzsguo: Liang Wenfeng:
Open source is actually good for AI commercialization, even if that sounds counteri...
- @Michaelzsguo: A large part of X users follows AI and open models, yet I've seen surprisingly little discussion of ...
- @teortaxesTex: interesting aspect here is that when Liang talks about talent and DeepSeek being "just ordinary peop...
- @Michaelzsguo: Liang Wenfeng was the first to understand that GPUs are more valuable than money sitting in the bank...
- @zephyr_z9: Production volume for 950DT is around 400k this year\n\nQT @deredleritt3r: Key quotes from the Liang...
8. firstadopter (Group Score: 207.2 | Individual: 31.4)
Cluster: 10 tweets | Engagement: 294 (Avg: 88) | Type: Tech
"Google’s strong cloud numbers belie a critical strategic mistake. It is mortgaging its future to goose current financials."
"It inflates today’s numbers while eroding tomorrow’s moat. Boosting the stock price now is myopic, short-term thinking when doing so will enable frontier AI startups to disrupt Google over time."
"Google is now supplying both OpenAI and Anthropic with inference capacity and underlying AI compute, giving oxygen to its own existential threats. History repeats itself. You should not help your main rivals with the most critical technology of the generation, when your own employees are starving for compute. It’s penny wise and pound foolish."\n\nQT @firstadopter: Google Is a Secular Short https://t.co/gYqTUFh50D
See 9 related tweets
- @MilkRoadAI: RT @KyleReidhead: Google just posted one of the best quarters in its history and the market sold $GO...
- @StockSavvyShay: $GOOGL is approaching its largest drawdown of the past year 😳 https://t.co/X6jEv9a04L\n\nQT @StockSa...
- @MTSlive: SITUATION EXPLAINED: Google just raised its 2026 AI spending forecast to $205 billion.
• Alphabet r...
- @ttunguz: Google Cloud grew 82% year-over-year to 22.3b consensus. The number ...
- @PhilipJohnston: They’re gona need a loooot of compute\n\nQT @chetanp: Google Cloud is now at ~$100B revenue run-rate...
9. GergelyOrosz (Group Score: 200.0 | Individual: 58.3)
Cluster: 5 tweets | Engagement: 1634 (Avg: 222) | Type: Tech
So let me get this straight: a car maker taking apart a rival's car to develop their new model is fine (Ford did this with Tesla and Chinese EVs)
But an AI company inspecting another AI company's model via prompting and inspecting outputs is a "distillation attack" and not fine?\n\nQT @mkratsios47: We have information that Moonshot AI distilled Anthropic’s Fable for the development of its K3 model.
To do this they developed a sophisticated internal platform to conduct large scale distillation against U.S. models, allowing them to quickly switch between multiple methods of access to avoid detection. Moonshot AI has also acquired GB300-equipped servers and has accessed GB300s in Thailand, likely to train its AI models. The United States strongly supports the free and fair development of AI, including a thriving competitive ecosystem that spans frontier models, specialized systems, open-source frameworks, and open-weight models. Legitimate AI distillation used to create smaller, more efficient models plays a vital role in this open innovation ecosystem. However, large-scale, covert industrial distillation aimed at stealing proprietary U.S. technology and undermining American research is unacceptable.
See 4 related tweets
- @RnaudBertrand: For anyone who knows about AI, this makes very little sense: the training run of a large-scale front...
- @quxiaoyin: Your (suggested) policy is not penalizing Kimi: Kimi will continue to distill Anthropic and launch b...
- @BlancheMinerva: 1. The timeline here makes this obviously impossible. Anyone even vaguely knowledgeable about AI dev...
- @WesRoth: according to the allegation, this was not a few developers manually asking Claude questions.
Moonsh...
10. jason (Group Score: 189.3 | Individual: 36.9)
Cluster: 9 tweets | Engagement: 2340 (Avg: 539) | Type: Tech
This fix is in!!!
Anthropic and OpenAI have bought protection from @realDonaldTrump — it seems!
Insane that after they both stole IP they’re looking to shut down access the free open source models that stole from them
Regulatory capture worked — and they did it in plain sight!!!
5% of their stock and $100m in lobbyists!
THIS ISNT A DEMOCRACY!!!\n\nQT @SecScottBessent: We support open-source AI and the innovation it unlocks. But open source is not open season on American IP. When PRC firms conduct covert, industrial-scale distillation attacks that cross the line into IP theft, sanctions and Entity List designations will be on the table.
See 8 related tweets
- @deliprao: The plan is
- ban Chinese AI till OpenAI/Anthropic IPO
- after IPO lock-in finishes and investors/fo...
- @_FORAB: 连线杂志爆料称，当前特朗普政府内部，就应对中国 AI 产生了分歧，白宫部分官员主张加强限制，并研究如何防止它们的开源行为，挤占美国企业客户份额。
但商务部则认为这些限制措施难以执行，不如推动美国头部...
- @jason: Would be a shame if the Chinese stole your Frontier Lab. https://t.co/ZEiVZMoYcn\n\nQT @SecScottBess...
- @curiouswavefn: So basically you will ensure that only cyber attackers but not American cyber defenders will have ac...
- @WSJ: OpenAI and Anthropic executives are sounding the alarm about the rise of cheap AI, particularly powe...
11. levelsio (Group Score: 173.7 | Individual: 53.8)
Cluster: 7 tweets | Engagement: 4493 (Avg: 691) | Type: Tech
🇪🇺 This year the European Union will again make more money from fining US tech companies
Than from the total tax income from Europe's own public tech companies! https://t.co/OTEzMJGfFV\n\nQT @EU_Commission: ⚠️ We have fined Google €890 million for breaching the Digital Markets Act, by promoting their own services and restricting customer access to alternatives on Google Search and Google Play.
https://t.co/zFChIQxKua https://t.co/QTqdpS6dyY
See 6 related tweets
- @levelsio: Crazier is that EU fines for US tech companies are now 2% of the total EU budget income!\n\nQT @leve...
- @tunguz: OMG
American tech mind cannot comprehend this.\n\nQT @levelsio: 🇪🇺 This year the European Union wil...
- @CNBC: European regulators have fined Google 890 million euros ($1 billion), alleging the company gives pr...
- @Cointelegraph: 🇪🇺 NEW: EU fines Google $1.02B under the Digital Markets Act for favoring its own services and restr...
- @FT: EU fines Google €890mn in test of Trump’s threats to protect Big Tech https://t.co/ATqNyOsmOC...
12. gabriel1 (Group Score: 161.2 | Individual: 34.9)
Cluster: 6 tweets | Engagement: 724 (Avg: 605) | Type: Tech
jane street and sk hynix 💀💀
etched building the final boss of inference\n\nQT @Etched: We’ve raised 10.3B valuation from Sequoia, Andreessen Horowitz, Jane Street, Argo, and SK Hynix.
Our mission is to run the world's inference. This round accelerates production of our inference clusters.
We've opened an 80,000-sqft, 10-MW facility 15 minutes from our office to expedite production and prototyping.
See 5 related tweets
- @sequoia: RT @Etched: We’ve raised 10.3B valuation from Sequoia, Andreessen Ho...
- @ypatil125: Incredible! @robertwachen and @UbertiGavin are building a generational company. Just banger executio...
- @_rockt: Insane ramp to build 10x more efficient inference. Congrats @robertwachen and @Etched team!\n\nQT @E...
- @mattshumer_: Etched has raised another round!
Proud to be an early investor.\n\nQT @Etched: We’ve raised $300M i...
- @richardczl: Congratulations! @carterjwilcox @robertwachen this is fire!\n\nQT @Etched: We’ve raised $300M in Ser...
13. rohanpaul_ai (Group Score: 154.1 | Individual: 37.2)
Cluster: 5 tweets | Engagement: 72 (Avg: 43) | Type: Tech
Two forward deployed engineers rebuilt a live FedEx supplier's delivery platform in 3.5 months. The original estimate was eight.
Their advantage did not begin with faster code generation, but with a repository-wide knowledge graph covering modules, dependencies, data flows, and domain language.
2 years ago you'd need 5 engineers and double the timeline for this output.
They mapped the entire repository into markdown files covering modules, dependencies, and data flows.
Agents read that map first, so generated code matched the real architecture.
That structure kept every task inside a context window the model could hold.
Humans reviewed all 122 merged pull requests, and the models wrote about 90% of the code.
You set up that infrastructure and two engineers produce what used to take a full team.
Kudos to @mardehaym’s Limestone Digital team.\n\nQT @mardehaym: A delivery orchestration platform that supplies FedEx, estimated 7 to 8 months to rebuild their core system.
We delivered in 3.5 months with 2 engineers and 122 merged pull requests in the first 90 days.
I want to walk through how, because "we used AI and it was faster" doesn't help anyone.
Their codebase had real users, real logistics operations, and real consequences if something broke.
This is brownfield, not a weekend prototype demo.
Week one, we didn't write code. We scanned the entire repo and built a knowledge graph: markdown files documenting every module, dependency, data flow, and domain term. Our agents read that graph before touching anything.
Teams skip this constantly. They point an agent at a repo, hand it a ticket, and get code that compiles but misunderstands the architecture. I've seen it across 100+ engagements. When someone says "AI doesn't work on our codebase," they almost always skipped context acquisition.
Every ticket ran a six-step loop: define, spec, plan, implement, test, document. The agent handles about 90% of code generation. A senior engineer reviews every PR, and nothing merges unless they can verify it, explain it, and debug it without the agent. We call that the V.U.E. gate.
The 50% time reduction came from three things: upfront context eliminated false starts, the agent handled mechanical coding so engineers focused on design and review, and tight spec-to-PR discipline kept scope from drifting.
AI compute: about $200 per developer per month.
Their CTO called our pod the "top performing team" inside the company. Both engineers got discretionary bonuses, twice.
That's what our Velocity Framework was built for.
See 4 related tweets
- @rohanpaul_ai: RT @mardehaym: A delivery orchestration platform that supplies FedEx, estimated 7 to 8 months to reb...
- @alex_prompter: If you're in private equity, this engineering team is what you need for your portco\n\nQT @mardehaym...
- @alex_prompter: "2 years ago you'd need 5 engineers and double the timeline for this output."
we live in insane tim...
- @mardehaym: RT @mardehaym: The 5-step loop that turns a board ticket into a reviewed PR. No developer writes cod...
14. wallstengine (Group Score: 145.0 | Individual: 31.3)
Cluster: 7 tweets | Engagement: 31 (Avg: 135) | Type: Tech
STRIPE WEIGHS $10B DEAL FOR OPENROUTER
Stripe is in talks to acquire OpenRouter in a deal that could value the AI-model marketplace at roughly $10 billion, according to The WSJ.
An agreement could be announced soon, although the talks may still fall apart or attract another buyer.
OpenRouter gives more than five million developers access to hundreds of models from OpenAI, Anthropic and open-weight providers through one interface. Customers can compare models, switch providers and route requests without building a separate integration for each company.
The reported price is notable. OpenRouter was valued at 10 billion sale would be nearly eight times that valuation only months later.
Stripe already handles OpenRouter's payments, invoicing, tax collection and fraud tools. Buying the company would move Stripe beyond processing AI revenue and into the infrastructure developers use to access, compare and pay for models.
See 6 related tweets
- @alex: RT @pitdesi: Interesting nuggets in @theinformation article on Stripe:
-They are a huge beneficiary...
- @amir: as @theinformation reported Wednesday:
"OpenRouter could be a good fit for Stripe: OpenRouter coll...
- @IPONewsroom_: STRIPE IS IN TALKS TO BUY AI MODEL MARKETPLACE OPENROUTER FOR AROUND $10 BILLION
Two months ago, in...
- @WSJTech: Stripe is in talks to acquire OpenRouter, a buzzy startup that helps developers choose between artif...
- @IPONewsroom_: Stripe is in talks to acquire OpenRouter, a startup that helps developers choose between AI models -...
15. firstadopter (Group Score: 137.6 | Individual: 41.1)
Cluster: 5 tweets | Engagement: 280 (Avg: 88) | Type: Tech
OK, here's the Google setup right now.
I expect Alphabet shares to fall much further. There's no valuation support here. Just as the multiple re-rated higher on Gemini 3 Pro reaching SOTA, it will now re-rate lower on Sundar's incompetent rambling answer to why Google is falling behind in frontier AI and the AI model progress uncertainty.
Gemini 3.5 Pro is obviously a bad model given the look ahead to Gemini 4.
Gemini 4 is the key question and likely many months away. I put the odds of Gemini 4 disappointing at 50%, given all the key AI talent departures. If it does disappoint, the stock will tank again.
If Gemini 4 is good and SOTA, you can re-buy right then at a better risk-reward setup.
Till then, the multiple compresses on uncertainty for the next few months.\n\nQT @firstadopter: Google Is a Secular Short https://t.co/gYqTUFh50D
See 4 related tweets
- @cryptopunk7213: yikes. google down -7% this morning despite beating earnings because they've fallen behind in AI
pe...
- @WesRoth: this growth requires enormous spending.
Alphabet expects to invest as much as $205 billion this yea...
- @0xrwu: Still boggles the mind that Google Search, with a $200B run-rate, is still growing at nearly 20% YoY...
- @moneycontrolcom: #Business | Is Google falling behind in AI? Sundar Pichai points to Gemini 4 and near-monthly model ...
16. Teslaconomics (Group Score: 137.3 | Individual: 35.5)
Cluster: 4 tweets | Engagement: 746 (Avg: 587) | Type: Tech
So… with a show of hands, who’s buying the fuckin dip? 🙋🏻♂️
$TSLA keeps handing out opportunities to the ones who can see the forest through the trees.
The market and short-term minded investors seem to be obsessing over one quarter, while I’m looking at where this company is going over the next 5-10 years.
Bro… this is what I see:
• Deliveries are up +25% YoY to a RECORD second quarter. • Revenue is up +26% YoY to 100B in trailing 12-month revenue for the FIRST TIME EVER. • FSD subscriptions up +56% YoY to 1.48 million, with a record 55%+ attach rate on new North American Tesla deliveries. • Energy storage deployments up 41% YoY, with Megapack demand continuing to grow. • Services revenue up 50% YoY, becoming a larger, higher-margin part of the Tesla’s business. • Robotaxis are now operating across seven major metros, with Cybercab production officially underway. • Optimus production lines are being installed this year. • And despite investing more aggressively than ever, Tesla still finished the quarter with $43.5 billion in cash.
Sure, margins were under pressure. And yes, free cash flow was negative.
But this is bc Tesla is spending $ billions building AI infrastructure, expanding factories, ramping Cybercab, Optimus, batteries, and compute. The Tesla team is investing to build and be the leader for the next decade.
Always remember, the biggest returns rarely come from buying when everything feels safe… they come from buying when everyone seems to be shitting their pants, while the fundamentals remain intact.
NFA, of course… it’s just my two cents from a nobody.
But I’m bullish then ever on the future of Tesla.
See 3 related tweets
- @Teslaconomics: Cash is the lifeblood of a business.
And if you believe that to be true, Tesla is doing just fine. ...
- @teslaownersSV: The signal has never been more bullish on $TSLA
Tesla just dropped Q2 2026 results: revenue crushe...
- @KyleReidhead: My thoughts on Tesla Q2 Earnings:
Let's start with the good news:
- record Q2 in deliveries and rev...
17. MatthewBerman (Group Score: 136.3 | Individual: 37.2)
Cluster: 6 tweets | Engagement: 483 (Avg: 181) | Type: Tech
Didn't this launch like a year ago?\n\nQT @OpenAI: Health in ChatGPT is starting to roll out to U.S. users.
You can securely connect Apple Health and supported medical records to understand your information in context, track what has changed, and have more informed conversations.
See 5 related tweets
- @tunguz: OpenAI just solved medicine.\n\nQT @OpenAI: Health in ChatGPT is starting to roll out to U.S. users....
- @testingcatalog: OpenAI started a broader rollout of ChatGPT Health for US users.
> ChatGPT can draw on your con...
- @Hesamation: OpenAI killed a million startups with one tweet. https://t.co/yaYIemsPJl\n\nQT @OpenAI: Health in Ch...
- @aiedge_: Sam Altman casually killing another 100+ startups https://t.co/fudkIekrw3\n\nQT @OpenAI: Health in C...
- @DataChaz: health startups after reading this: https://t.co/ilxLWe2usA\n\nQT @OpenAI: Health in ChatGPT is star...
18. ruima (Group Score: 133.6 | Individual: 29.4)
Cluster: 7 tweets | Engagement: 596 (Avg: 122) | Type: Tech
RT @SophiaCai99: News: Nearly 200 Silicon Valley companies, including Proton and Y-Combinator, are urging the Trump administration not to cut off access to Chinese open-weight AI models or risk crippling the next generation of U.S. startups. https://t.co/Cj30cyeonb
See 6 related tweets
- @DadaJudith: RT @MTSlive: SITUATION UPDATE: The Trump administration is split on how to handle Chinese open sourc...
- @quxiaoyin: this is why U.S has the best startup ecosystem in the world. it can self-correct fast\n\nQT @SophiaC...
- @PeterDiamandis: Here's the thing, OpenAI, Anthropic and Google are each spending tens of billions of dollars in aggr...
- @pstAsiatech: Almost 200 Silicon Valley companies, including Proton and Y Combinator, are urging the Trump adminis...
- @pstAsiatech: Why the US is losing Chinese AI stars More entrepreneurs from China believe there are now greater o...
19. m_sirovatka (Group Score: 124.1 | Individual: 29.7)
Cluster: 5 tweets | Engagement: 131 (Avg: 71) | Type: Tech
I’m gonna be honest, no matter how hard I try, the talk is not gonna be better than this thumbnail, be there today\n\nQT @RedHat_AI: vLLM Office Hours today at 2pm ET: RL at 1T Scale, a prime-rl performance deep dive with @m_sirovatka (@PrimeIntellect). Training trillion-parameter MoE models like GLM-5.1 on agentic RL, plus what's new in vLLM 0.25 by @mgoin_. Get a recurring cal invite: https://t.co/KkF6DfGs1q
See 4 related tweets
- @vllm_project: What does trillion-scale agentic RL look like on the inference side?
@PrimeIntellect's prime-rl 0.6...
- @samsja19: prime researcher are c̶r̶a̶c̶k̶e̶d̶ cat\n\nQT @RedHat_AI: vLLM Office Hours today at 2pm ET: RL at...
- @eliebakouch: inference cat supremacy https://t.co/EIjsMqhvY8\n\nQT @RedHat_AI: vLLM Office Hours today at 2pm ET:...
- @willccbb: RT @RedHat_AI: vLLM Office Hours today at 2pm ET: RL at 1T Scale, a prime-rl performance deep dive w...
20. Breaking911 (Group Score: 123.3 | Individual: 27.3)
Cluster: 6 tweets | Engagement: 649 (Avg: 1329) | Type: Tech
PRESIDENT TRUMP: Data centers and AI are dramatically increasing the demand for electricity. It's only fair that the cost of building the new infrastructure required to meet this demand be borne by the corporations — not by American consumers. Under the Ratepayer Protection Pledge, America's largest tech companies have formally committed fund or build all energy infrastructure they are placing on the grid.
See 5 related tweets
- @FoxNews: NEW: President Trump announces major progress on powering American tech, emphasizing that massive ne...
- @Polymarket: JUST IN: Trump to expand his “Ratepayer Protection Pledge,” aimed at ensuring AI companies cover ele...
- @FoxNews: “I hope you remember me, because what we've done is something that nobody thought was possible.”
Pr...
- @MarioNawfal: 🇺🇸 Trump just said AI data center companies will have to produce MORE energy than they use.
The exc...
- @Techmeme: Trump says 200 additional stakeholders, including major utilities, joined a voluntary pledge to prot...