- Published on
科技热门推文——2026年8月5日
- Authors
- Name
- geeknotes
今日科技动态：人工智能在汽车、终端设备和企业工作流等领域持续取得进展。英伟达发布自动驾驶推理模型，Liquid AI 推出端侧智能体，Cursor 则将业务拓展至 Google Workspace。由提示词驱动的应用开发和广告工具，进一步凸显了向自然交互界面转型的趋势。与此同时，算力需求激增，以及美国可能限制在华数据中心使用相关组件，使基础设施问题受到更多关注。此外，据报道，Airtable 的收购交易反映出 SaaS 行业整合加速，而一个迅速传播的 npm 蠕虫病毒则凸显了日益严峻的软件供应链风险。
1. genalewislaw (Group Score: 329.9 | Individual: 48.2)
Cluster: 13 tweets | Engagement: 2122 (Avg: 203) | Type: Tech
RT @JensenHuang: Today, we’re launching Alpamayo 2 Super, our frontier open reasoning model for autonomous vehicles.
Beyond seeing, Alpamayo understands and reasons through the complex world - thinks before it acts.
It’s a powerful backbone for robotaxis, trucks, shuttles, delivery vans, tractors and the long tail of mobile robots—billions of autonomous machines someday.
We’re releasing it for commercial use under OpenMDW-1.1 so teams can inspect it, fine-tune it and deploy it—open models advance safety and security.
The next wave of AI is robotics—and it starts with autonomous vehicles.
Great work, Alpamayo team!
See 12 related tweets
- @VaibhavSisinty: NVIDIA isn’t building the next Tesla.
It’s building the brain for the next 100 Teslas.
Today, NVID...
- @wallstengine: NVIDIA has released Alpamayo 2 Super, an open-weight reasoning system built for robotaxis, trucks an...
- @MTSlive: SITUATION EXPLAINED: Nvidia opened Alpamayo 2 Super for commercial use, its frontier reasoning model...
- @damianplayer: the most valuable company on earth is launching frontier models out of an 11 day old X account.
leg...
- @StockMKTNewz: Nvidia $NVDA CEO Jensen Huang just posted this:
“Today, we’re launching Alpamayo 2 Super, our front...
2. wallstengine (Group Score: 305.8 | Individual: 35.4)
Cluster: 10 tweets | Engagement: 52 (Avg: 132) | Type: Tech
In the AI era, the programming language is human.
@boltdotnew has launched working, full-stack business app templates built by professional designers and customizable with prompts.
When everyone can program, knowing what to program becomes the real advantage. https://t.co/tgphtGO43r\n\nQT @boltdotnew: A truly standout website costs tens of thousands of dollars in design.
So we did it for you, with a handful of the world's best designers and agencies. For free.
Introducing the Bolt Template Marketplace: real, full stack apps you can open in Bolt and make yours 🧵 https://t.co/YUFVP3wzZv
See 9 related tweets
- @StockSavvyShay: Bolt is a great example of the new AI application business where you can rent intelligence from mode...
- @Mayhem4Markets: you can always spot an AI-built app from across the room.
same gradient. same rounded cards. same t...
- @haider1: last week i partnered with @boltdotnew to try their new template marketplace
instead of starting fr...
- @exec_sum: NEWS: Bolt launches its first website & app template marketplace.
@boltdotnew launched a library of...
- @VadimStrizheus: if you’re a vibe-coding beginner and you don’t know where to start, read this:
one of the best way...
3. cryptopunk7213 (Group Score: 285.6 | Individual: 36.7)
Cluster: 16 tweets | Engagement: 353 (Avg: 103) | Type: Tech
next GPT models are going to be too large for laptops alone to run
codex cloud becomes a bedrock for running people’s ai agents 24/7. openai will make a killing from rent compute.
models are going to need so much more memory and 2027 capacity just sold out.
we are accelerating a lot quicker than i thought\n\nQT @thsottiaux: Given some of the results I'm seeing recently, it's pretty clear Codex is a good harness.
But it will seem primitive in 2-3 months and we're about to go through another major evolution in how we use AI at the frontier. The next generation of models need more than your laptop.
See 15 related tweets
- @haider1: OpenAI: "the next generation of models needs more than your laptop"
which suggests that, over the n...
- @AndrewCurran_: 'we're about to go through another major evolution in how we use AI at the frontier'\n\nQT @thsottia...
- @kimmonismus: OpenAI: "The next generation of models need more than your laptop."
"But it will seem primitive in ...
- @mark_k: Tibo of @OpenAI Codex wrote something interesting here:
"The next generation of models need more th...
- @zephyr_z9: "The next generation of models need more than your laptop."\n\nQT @thsottiaux: Given some of the res...
4. svpino (Group Score: 231.1 | Individual: 28.9)
Cluster: 13 tweets | Engagement: 93 (Avg: 159) | Type: Tech
Calling the cheapest possible model is the wrong strategy. Instead, you should optimize for the costs of completing a task.
When optimizing an agent, you should account for:
• Reasoning effort • Retries • Cache state • Downstream impact of each routing decision
For this to work, you need to make routing decisions across the full session, not one request at a time.
Not Diamond Code is doing exactly this: selecting model + reasoning effort at each step. They can maintain frontier-level quality responses at around 20% lower inference costs.\n\nQT @tomas_hk: Today we’re announcing Not Diamond Code, the world’s most powerful intelligent model router for long-horizon coding agents.
Not Diamond works with any gateway or harness, including Claude Code, to select the best model and reasoning effort for each step, reducing costs by 20-65% without impacting quality.
See 12 related tweets
- @wallstengine: NOT DIAMOND LAUNCHED AN INTELLIGENT MODEL ROUTER FOR CODING AGENTS
Not Diamond Code supports Claude...
- @Mayhem4Markets: Your coding agent just used Opus at max reasoning to list files in a directory. That's the kind of t...
- @Meer_AIIT: most teams trying to cut coding agent spend start by swapping to a cheaper model mid session.
what ...
- @damianplayer: bullish on finished tasks.\n\nQT @tomas_hk: Today we’re announcing Not Diamond Code, the world’s mos...
- @VadimStrizheus: Not Diamond Code is making a pretty wild claim
20–65% lower costs for coding agents without a quali...
5. BullTheoryio (Group Score: 219.5 | Individual: 28.0)
Cluster: 12 tweets | Engagement: 358 (Avg: 1058) | Type: Tech
BREAKING: The Trump administration is drafting a ban on imports of Chinese data center components, per Reuters.
The FCC is working on a measure to bar new Chinese optical transceivers. These move data over fiber optic cables inside data centers and sit at the core of every AI facility being built right now.
Officials want it published this year, taking effect immediately.
The concern is data theft, malware and service disruption at the facilities running US AI models.
The administration wants to act before Chinese hardware gets embedded the way Huawei equipment did in US telecom networks, which took years and billions to remove.
The Chinese embassy said Beijing "will take all necessary measures in response to any action that causes material harm to its interests."
See 11 related tweets
- @wallstengine: TRUMP ADMINISTRATION DRAFTS BAN ON CHINESE DATA CENTER COMPONENTS - REUTERS
The FCC is preparing a ...
- @Crypto_Jargon: 🚨 THE US IS ABOUT TO BAN CHINESE COMPONENTS INSIDE EVERY AI DATA CENTER.
Optical transceivers are t...
- @norveclifinance: The market is reacting to headlines, not reality.
A broad ban on Chinese data center components wou...
- @StockSavvyShay: The Trump administration is drafting a ban on new Chinese optical transceivers which is a critical n...
- @DeItaone: U.S. PLANS BAN ON NEW CHINESE DATA CENTER PARTS
Reuters reports the Trump administration is draftin...
6. FellMentKE (Group Score: 212.0 | Individual: 33.1)
Cluster: 7 tweets | Engagement: 140 (Avg: 170) | Type: Tech
I don't think the next great AI interface will be a better prompt box.
It may be the world already in front of us.
We don't naturally experience life by typing questions. We glance at a room, open our closet, sit down for lunch, or walk past a plant. The most useful AI won't wait for us to translate those moments into words. It will understand the context as it happens.
That's why I find NewEyes interesting. It's one of the first apps using AI directly on your device to understand what your camera sees, turning the camera from something that merely records life into an intelligence layer that remembers what matters and helps in the moment.
Styling an outfit, rethinking a room, tracking a meal, or noticing when a plant needs attention.
The shift isn't just from chat to vision. It's from asking AI for help to having intelligence quietly work alongside you.\n\nQT @CollovLabs: What if your camera could…
🛍️ Shop what you see. 👗 Style what you wear. 🛋️ Design the room around you. 🍽️ Count the calories on your plate. 🪴 Remind you when your plants are thirsty.
And that's just the beginning. With NewEyes, your camera doesn't just capture the world — it understands it, remembers it, and acts on it. Working for you before you even ask.
One of the first consumer AI apps powered by on-device multimodal intelligence.
A visual intelligence layer, built for every camera.
- Real-time perception.
- Persistent memory.
- Long-horizon execution.
- Proactive, not reactive.
NewEyes. Personal intelligence with eyes.
📱Download: https://t.co/D4E3Ao3CEL
See 6 related tweets
- @ai_explorer25: Trained on the entire internet, your AI still can't see the room it's in.
NewEyes delivers a camer...
- @kimmonismus: This is cool: Point your phone at a Japanese menu and dinner is handled.
That's the demo Collov Lab...
- @svpino: The idea that AI can understand and act on what you are looking at is pretty interesting.
Not "desc...
- @AngryTomtweets: Cameras are becoming intelligent.
NewEyes doesn’t just see the world around you. It understands it ...
- @TheoBuildsAI: Google Lens recognizes. ChatGPT describes. NewEyes actually acts.
That’s the shift that stood out t...
7. deedydas (Group Score: 209.3 | Individual: 41.9)
Cluster: 9 tweets | Engagement: 3426 (Avg: 528) | Type: Tech
Wow. Airtable, founded in 2012 and once valued at $11.7B, is getting acquired by Bending Spoons, founded 2013, at 2.7x ARR.
A once hot startup; now an unfortunate victim of the SaaS bust. It raised 1.285B EV ($2.25B equity value), just clearing its preference stack.
That implies common and early holders split the remaining ~1.285B! 😍 https://t.co/TsVlDOMX26
See 8 related tweets
- @wallstengine: BENDING SPOONS TO ACQUIRE AIRTABLE FOR $1.28B
Bending Spoons, an Italian software company known for...
- @rvivek: Congrats @howietl! Everyone who's dissing on this is missing the big picture.
-- The total value o...
- @exec_sum: NEWS: Italian app investor Bending Spoons acquired Airtable in a $1.3B cash deal
This is Bending S...
- @zerohedge: RT @Restructuring__: Holy shit this is really bad (and honestly shocking)
Airtable raised at $11.7B...
- @exec_sum: NEWS: Italian app investor Bending Spoons acquired Airtable for $1.3B
This is ~80% lower than Airta...
8. Parul_Gautam7 (Group Score: 204.1 | Individual: 39.8)
Cluster: 6 tweets | Engagement: 416 (Avg: 183) | Type: Tech
The most interesting part isn't that Arcads generates ads.
It's that it starts by understanding your website, brand, and audience before creating them.
That's a much more practical workflow than starting from a blank page. https://t.co/vwSjTzvHqt\n\nQT @arcads_ai: We raised $16M to build Claude Code for Meta Ads
Introducing Mark: The only AI built for winning ads
10,000 of the best performance marketing teams use it to scale their ads!
It has persistent brand memory, instant access to your entire assets library, and is trained on millions of winning ads.
Mark can also:
Create hyper-realistic UGC Design motion graphics Localize campaigns worldwide
Everything a creative team does. Faster. Better. Autonomous.
Today we're releasing it to all Arcads users
See 5 related tweets
- @rom1trs: AI is running our distribution now: https://t.co/AZRDBj8V05\n\nQT @arcads_ai: We raised $16M to buil...
- @VadimStrizheus: meta wants to fully automate ad creation by the end of 2026
arcads shipped it first 💀
give it a we...
- @aakashgupta: Apple's ATT broke precision ad targeting in 2021. Five years later, winning on paid comes down to cr...
- @heyrobinai: mark zuckerberg said he'd automate all advertising by 2027
Arcads did it in 2026 and just dropped A...
- @cgtwts: this is claude for marketing and it is genuinely terrifying\n\nQT @arcads_ai: We raised $16M to buil...
9. heyrobinai (Group Score: 173.6 | Individual: 32.1)
Cluster: 7 tweets | Engagement: 12 (Avg: 61) | Type: Tech
hollywood legends are taking AI this seriously
Pixar co-founder Edwin Catmull is literally judging AI filmmakers right now.. the guy who invented 3D graphics.. 5 Oscars on his shelf
he's on the jury for Higgsfield's Global Film Festival:
500k for first place submissions open Aug 7, close Aug 31 any genre, 3 min minimum, solo or a team of 4 open worldwide
24 days to make a film.
if that doesn't tell you where cinema is going.. nothing will\n\nQT @higgsfield_ai: Pixar co-founder Edwin Catmull joins the Higgsfield Global Film Festival as a jury member.
5x Oscar Winner. Ex-President of Walt Disney Animation Studios. Executive producer of Toy Story. Turing Award laureate.
The man who invented 3D graphics will judge the first generation of AI filmmakers.
See 6 related tweets
- @AnatoliKopadze: this is genuinely hard to believe
we're going to have an AI film festival with $1M in prize pool
o...
- @higgsfield: Pixar co-founder Edwin Catmull joins the Higgsfield Global Film Festival as a jury member.\n\nQT @hi...
- @VadimStrizheus: bro hold up….do you understand what just happened?😭🥀
the co-founder of Pixar just joined the $1M Hi...
- @DataChaz: 5 Oscars. 1 Turing Award. Pixar co-founder. Former President of Walt Disney Animation Studios.
Now ...
- @cgtwts: The person who helped build the last major shift in animation is now judging the next one
Edwin Cat...
10. BlancheMinerva (Group Score: 172.7 | Individual: 27.7)
Cluster: 7 tweets | Engagement: 12 (Avg: 69) | Type: Tech
Backroom deals for how the threat of advanced AI models are going to be judged is anti-democratic and makes the world a more dangerous place.\n\nQT @axios: NEW: White House does not plan to publicly release its new framework for evaluating advanced AI models.
See 6 related tweets
- @theobearman: This is a big disappointment. If nothing else, the public now won’t know whether the framework was e...
- @coinbureau: ⚡️JUST IN: The White House will keep its new AI evaluation framework private, limiting access to sel...
- @MTSlive: SITUATION UPDATE: The White House has completed its voluntary AI framework for evaluating advanced m...
- @m_ccuri: RT @axios: NEW: White House does not plan to publicly release its new framework for evaluating advan...
- @theojaffee: https://t.co/uVW6TO0Lyq\n\nQT @MTSlive: SITUATION UPDATE: The White House has completed its voluntar...
11. StockSavvyShay (Group Score: 164.1 | Individual: 35.8)
Cluster: 6 tweets | Engagement: 991 (Avg: 636) | Type: Tech
150 and is now up more than 20% today 😳 https://t.co/KZZr10SavQ\n\nQT @StockSavvyShay: $PLTR Rule of 40 score reached 155 and the metric has now expanded for 12 consecutive quarters as operating leverage continues improving with scale.
That means Palantir is growing close to 100% while converting more than half of revenue into operating income which is a combination almost unheard of in large-scale software.
AMZN, META are spending hundreds of billions building the AI factories while Palantir sits above that infrastructure and turns the resulting intelligence into real-world outcomes without funding the physical buildout allowing far more incremental revenue to flow into profit and cash.
See 5 related tweets
- @StockSavvyShay: Deutsche Bank upgrades 200 target calling Palantir a “time traveler” operating ...
- @StockSavvyShay: PLTR is one of the cl...
- @MilkRoadAI: RT @MilkRoadAI: Palantir just posted a Rule of 40 score of 155%, and that number alone should terrif...
- @danielnewmanUV: $PLTR dunked on the “End of Enterprise Software” crowd with its Rule of 155 growth and massive comme...
- @coinbureau: ⚡️UPDATE: Palantir $PLTR shares soared 30% after stellar Q2 earnings, driven by a 149% revenue surge...
12. AltimeterCap (Group Score: 163.6 | Individual: 34.9)
Cluster: 7 tweets | Engagement: 7 (Avg: 23) | Type: Tech
The AI era is creating one of the largest infrastructure buildouts in modern history.
As intelligence becomes foundational to every enterprise, the demand for compute, power, and AI infrastructure will continue to accelerate. We're excited to support some of the builders making that future possible.
Congratulations to Volta on emerging from stealth with a mission to build the Utility of Compute.
Altimeter Partner @jaminball led our investment, recognizing the opportunity to back a team building a fully vertically integrated AI infrastructure platform spanning capital, powered land, data centers, compute, software, and operations.
Congratulations to Ricard Boada, Sofia Gumuzio, and the entire Volta team. We look forward to what's ahead.\n\nQT @business: A new AI cloud company has raised $300 million in venture funding https://t.co/hFkXsHNkep
See 6 related tweets
- @a16z: We're thrilled to co-lead Volta's Series A.
Compute isn't distributed evenly. Most startups don't h...
- @AltimeterCap: RT @jaminball: Very excited to partner with Volta and join the board alongside @RaghuRaghuram / @sha...
- @Cointelegraph: 🔥 UPDATE: Anthropic has signed a $10B computing capacity deal with AI cloud startup Volta Infra Hold...
- @coinbureau: ⚡️JUST IN: Anthropic signs a $10 billion computing deal with AI cloud startup Volta Infra Holdings. ...
- @Dell: We're excited to support the NSF's new State and Regional AI Infrastructure Hubs program, an initiat...
13. robinebers (Group Score: 157.3 | Individual: 46.8)
Cluster: 5 tweets | Engagement: 628 (Avg: 59) | Type: Tech
RT @cursor_ai: Cursor can now read, write, and act across your Google Workspace.
New plugins give agents direct access to Gmail, Google Drive, Calendar, Docs, and Sheets. https://t.co/1GGQdthf7E
See 4 related tweets
- @yuhasbeentaken: Half the replies are asking what the use case is. From a growth engineer: pull campaign numbers from...
- @badlogicgames: so everybody is building chatboxes with connectors now.\n\nQT @cursor_ai: Cursor can now read, write...
- @agentnative_: Cursor, Codex and Claude Desktop.. The big 3 super apps.
Prediction: Cursor makes a hard marketing ...
- @mark_k: It looks like Cursor might expand beyond coding: Its new plugin enables access to your entire Google...
14. SocketSecurity (Group Score: 155.5 | Individual: 62.9)
Cluster: 4 tweets | Engagement: 1660 (Avg: 165) | Type: Tech
🚨 Update: Watching this npm worm propagate in real time, we’re now tracking 2,234 affected package artifacts across 444 unique packages, and it’s still spreading.
Average detection time: 5 min and 18 seconds after publication. Our campaign page includes all packages/versions.\n\nQT @SocketSecurity: 🚨 Active npm supply chain attack: keyv@6.0.0 and 13 other packages have been compromised. keyv alone gets 154M weekly downloads.
The worm steals cloud and CI credentials, then uses stolen npm tokens to publish trojanized versions of more packages. https://t.co/IFkfklQw7K
See 3 related tweets
- @KyeGomezB: RT @IntCyberDigest: ‼️ BREAKING: An active npm supply chain attack has compromised at least 868 pack...
- @SocketSecurity: RT @feross: 🚨 Active supply chain attack on npm: keyv and cacheable are compromised right now, and t...
- @10_X_eng: at this point pre-install and post install hooks are going to get ripped out.\n\nQT @SocketSecurity:...
15. tugot17 (Group Score: 153.3 | Individual: 31.5)
Cluster: 7 tweets | Engagement: 91 (Avg: 26) | Type: Tech
already fast and will get faster 😏 https://t.co/MlEzt9ivPk\n\nQT @liquidai: Today we release LFM2.5-2.6B, an agentic model that runs entirely on-device. It plans, calls tools, and works through multi-step tasks on phones, laptops, PCs, and robots. Data never leaves the device, and the marginal cost of each run is essentially zero.
Pre-trained on ~34T tokens LFM2.5 flagship hybrid architecture Context length: 128K Vocab size: 128K balanced intelligence per watt customizable on a single GPU for any specialized task LFM2 open-weight license
Comparable or better scores compared to models up to nearly 4x its size:
ToolSandbox 77.83, ahead of Qwen3.5-9B at 76.44 Multi-IF 80.07, ahead of Gemma-4-E4B-it at 77.35 IFStruct 85.49, ahead of Qwen3.5-9B at 78.50
🧵
See 6 related tweets
- @TeksEdge: 🆕 A new 2.6B Local AI model is beating Qwen3.5-9B on several key agent/tool benchmarks. Gotta try th...
- @victormustar: RT @liquidai: Today we release LFM2.5-2.6B, an agentic model that runs entirely on-device. It plans,...
- @sgl_project: SGLang has Day 0 support for @liquidai's latest LFM2.5-2.6B!
LFM2.5 hybrid architecture, ~34T pre-...
- @liquidai: RT @poly0015iew: LFM2.5-2.6B from Liquid AI - a 2.6B parameter model that outperforms models 3-4× la...
- @liquidai: RT @songdng: LFM2.5-2.6B is OUT!
I LOVED post-training this model, especially working on agentic R...
16. norveclifinance (Group Score: 141.8 | Individual: 34.4)
Cluster: 5 tweets | Engagement: 5 (Avg: 144) | Type: Tech
China is moving at a pace the market continues to underestimate.
Chinese memory makers are rapidly expanding capacity. They’re ramping up smartphone memory production while building new fabs to accelerate entry into other memory segments.
The AI and memory euphoria was built on the assumption that supply would stay tight for years. That assumption is being challenged.
In my view, China has already started deflating both the AI bubble and the memory bubble.
NVDA QCOM ASML Samsung #AI #Semiconductors #Memory #HBM #DRAM #NAND #Investing #Stocks #Borsa #Yatırım #Teknoloji #ChipStocks #WallStreet #China\n\nQT @Reuters: EXCLUSIVE: CXMT plans second chip plant in Beijing and is in talks on its funding, sources say https://t.co/OoYF2PGc83 https://t.co/OoYF2PGc83
See 4 related tweets
- @norveclifinance: China is moving at a pace the market continues to underestimate.
Chinese memory makers are rapidly ...
- @norveclifinance: China is moving at a pace the market continues to underestimate.
Chinese memory makers are rapidly ...
- @wallstengine: HP, ASUS AND ACER BEGIN LIMITED USE OF CXMT DRAM - NIKKEI
The PC makers have started using small vo...
- @norveclifinance: China is moving at an incredible pace. Their factories seem to go from blueprint to production at re...
17. yacineMTB (Group Score: 140.1 | Individual: 29.6)
Cluster: 7 tweets | Engagement: 1034 (Avg: 251) | Type: Tech
this is just absolutely brutal\n\nQT @ValsAI: DeepSeek V4 Flash (0731) is the cheapest model on the Vals Index to score above 60, and 35x cheaper than the next best model.
That efficiency comes almost entirely from coding. Three of its four strongest Index results are coding and agentic tasks, and the cost gap on those is wider still.
See 6 related tweets
- @EntelligenceAI: DeepSeek V4 Flash reportedly scores close to Opus 4.8 and GPT-5.3 Codex on LiveCodeBench.
On Vibe C...
- @EntelligenceAI: DeepSeek V4 Flash reportedly costs just $0.03 per WeirdML test (yeah, just 3 cents)
For comparison:...
- @zephyr_z9: DAMN Pro will be a little bit more expensive but performance will be super high\n\nQT @ValsAI: DeepS...
- @teortaxesTex: crushing superiority.\n\nQT @ValsAI: Vibe Code Bench is the clearest case. V4 Flash scores 74.7 at 2...
- @exec_sum: DeepSeek's latest AI model is 105x cheaper than Claude Fable 5
Their new V4-Flash model ranks on p...
18. patrick_oshag (Group Score: 136.1 | Individual: 31.1)
Cluster: 5 tweets | Engagement: 669 (Avg: 273) | Type: Tech
Gavin on why the hyperscalers are under-earning:
" I literally spoke to a company this morning who rented a cluster of several thousand Blackwells at somewhere in the mid $2 per GPU hour, and this is one of the sexiest startups that people want to be in business with.
They're renting the exact same cluster, and they're hoping 7 months later to pay just under $4.
That's pretty crazy because a really gentle decline in prices would be bullish. Instead, we're up, depending on the starting point, 50 to 60% in six or seven months.
And there have been so many anecdotes like that. One of the inference clouds, I think it was Baseten, they went on a podcast and essentially said, "We are planning to pay 100% more for Blackwells when our contract expires.
And that just means that essentially all the hyperscalers are under-earning.
My main mission out here this week is: tell me something negative."\n\nQT @patrick_oshag: My seventh conversation with @GavinSBaker.
It's about the gap between what the market is doing and what companies are seeing. It's been a tough month or so for public AI names, but there's no sign of a slowdown on the ground in Silicon Valley.
We discuss:
- Why old GPUs are getting more expensive
- Open source vs the frontier
- Claude as Wall Street's Walter Cronkite
- Whether the buildout gets funded out of cash flow or debt
- SpaceX as a data center company
- And more
Enjoy!
TIMESTAMPS 0:00 Intro 1:17 AI Selloff vs. Fundamentals 10:20 Financing the AI Buildout 18:18 GPU Prices Keep Rising 24:23 Claude Moves Markets 29:01 What Could Break the Thesis 36:59 The Memory Supply War 42:08 Nvidia’s New Playbook 50:14 China and Open Source 61:51 Data Centers and Regulation 71:11 SpaceX and Orbital Compute
See 4 related tweets
- @zephyr_z9: Gavin: "SSI says that they are going to come out with their model in August"\n\nQT @patrick_oshag: M...
- @patrick_oshag: "Have you met anyone that's more bullish than you?"
Gavin: "Everyone out here is more bullish than ...
- @patrick_oshag: Gavin on SpaceX’s compute ambitions:
“A Substack writer, @FundaAI, thinks that SpaceX is going to b...
- @bizbreakdowns: RT @patrick_oshag: My seventh conversation with @GavinSBaker.
It's about the gap between what the m...
19. wallstengine (Group Score: 135.9 | Individual: 36.1)
Cluster: 4 tweets | Engagement: 39 (Avg: 132) | Type: Tech
People are already charging 10K to help companies implement AI employees.
That work now has a name.
"vibe deploying", is getting paid to put AI employees to work inside real businesses. https://t.co/07s6hIq90I
Now Viktor has launched a specialist program in which you can:
Bring your clients and connect Viktor to their tools
Build workflows and help their teams get started
Set your implementation fee and keep 100%
Earn 20% of eligible referred revenue for 12 months
Receive 2,000 activation bonuses for qualifying new companies.
The model is only valuable when a business can use it. That implementation gap is becoming a market of its own.
@viktor_com\n\nQT @Antoni_Olendzki: vibecoding was for building. vibe deploying is for getting paid.
Installing AI employees at real businesses. No code. 10K per setup plus 20% of client spend for 12 months.
First guy in made six figures in 30 days (partner reported, results vary.)
@viktor_com https://t.co/usKVDBlwW8
See 3 related tweets
- @VadimStrizheus: Vibe deploying is getting paid to put AI employees to work inside real businesses.
Anthropic, Black...
- @svpino: A great opportunity for those who want to help deploy AI employees in real companies.
Viktor's Spec...
- @dani_avila7: Every major software platform eventually creates a services ecosystem around it
Vibe deploying is g...
20. WesRoth (Group Score: 130.5 | Individual: 36.9)
Cluster: 4 tweets | Engagement: 48 (Avg: 28) | Type: Tech
OpenAI is replacing the walkie-talkie model of voice AI with something much closer to a real conversation.
GPT-Live can listen while it speaks, removing the turn detector that previously had to guess when a user had finished talking.
That means fewer awkward pauses, fewer accidental interruptions, and the ability to handle overlapping speech more naturally.
When a question requires deeper reasoning, search, or tools, GPT-Live can quietly delegate the work to a frontier model such as GPT-5.5 while continuing the conversation itself.
OpenAI rebuilt the voice stack over six months, separating audio from slower tool calls and reducing session startup from six network round trips to one. The result is sub-second responsiveness at ChatGPT scale.
This infrastructure already powers ChatGPT’s ability to control a computer and coordinate agents through voice and it will underpin the upcoming GPT-Live API.\n\nQT @OpenAI: GPT-Live can listen while it speaks.
To make that feel natural at ChatGPT scale, we rebuilt the voice stack from client to model.
This new architecture keeps audio flowing continuously, so deeper reasoning and tool use don't interrupt the conversation. https://t.co/53VTTsqXg0
See 3 related tweets
- @rohanpaul_ai: OpenAI explained how it built GPT-Live for real-time voice conversations.
It's built around continu...
- @altryne: This is great, now we need more words than just "checking" or "checking that for you" every time voi...
- @rohanpaul_ai: RT @rohanpaul_ai: OpenAI explained how it built GPT-Live for real-time voice conversations.
It's bu...