科技热门推文——2026年8月17日

今日科技界，人工智能的潜力与风险成为讨论焦点。达里奥·阿莫迪预计，未来十年内医学领域有望在疾病治疗方面取得突破，但同时警告权力可能高度集中，由此引发了是否应放缓技术进步的争论。工程师们分享了新兴的人工智能智能体架构、技能图谱、分词器效率优化方案，以及 Grok Build 的远程控制能力，同时也提出了何种成果才算真正研究的问题。与此同时，超大规模云服务商的基础设施支出预计将大幅增长，欧盟信息披露规定即将实施；初创企业界人士则强调，加州投资市场展现出韧性，可规模化的获客渠道正在形成，人工智能驱动的新型商业模式也不断涌现。

1. rohanpaul_ai (Group Score: 865.8 | Individual: 57.7)

Cluster: 24 tweets | Engagement: 362 (Avg: 38) | Type: Tech

Dario Amodei broke his silence!

• Predics that AI could help cure most human disease within roughly 5–10 years.

• says AI is structurally prone to concentrating power, even without regulation.

• He wants regulation designed to slow frontier AI labs while giving smaller challengers more room to catch up.

• Argues open-weight AI does not solve concentration, because even with open-source serious capability still depends on scarce compute and chips.

• He supports pre-deployment testing of frontier models, with open-weight models tested too once they approach frontier capability

• Says AI companies will not win public trust through better marketing. They need to deliver undeniable real-world breakthroughs first.

So now he just tied public trust to visible medical outcomes of AI, looks like Anthropic’s biology push is for real.\n\nQT @DarioAmodei: 1/2 Thanks Gavin for an especially thoughtful exchange. I don't usually spend much time on social media but I wanted to engage here because it really brings out the heart of an important conversation.

First, on regulation, I think that “either concentrate it in the hands of a chosen few companies and politicians via regulation or distribute it widely” is a false choice.  I know that there’s a sort of Silicon Valley shorthand where regulation = regulatory capture = concentration of power, but I’ve always found this to be an overly simplified picture of the world.  Many people outside this bubble think of regulation as something that constrains corporate power and benefits ordinary people.  I don’t necessarily agree with that perspective either, rather I think it’s complicated and really depends on what the “regulation” consists of.  But in particular I think that those in the “regulation = regulatory capture = concentration of power” frame often underrate the decentralizing power of objective and fair institutional processes.  A crude analogy is that the formal court system can sometimes feel stuffy and elitist, but it does a much better job of defending the rights of vulnerable individuals than the alternative, mob justice.  At their best, institutions can vest power in ideas rather than people, and thereby decentralize that power.

This is why Anthropic has always made its policy proposals very carefully.  We try very hard to make proposals that disadvantage (slow down) frontier AI companies while advantaging smaller competitors.  California’s SB53 (which we supported), and even the much-maligned SB 1047 (which we were ambivalent on), completely exempt any company below a certain amount of revenue or model training costs from being covered at all (it was $500M for SB 53, lower for 1047 but we objected to that).  More recently the testing process we’ve advocated for at CAISI and the White House involves more rigorous tests for frontier models than off-frontier models — something that differentially advantages challengers.  Similarly, the “Pacing the Frontier” letter envisions (or at least Anthropic’s preferred implementation of it envisions) modulating the pace of the very best models while not constraining those who are catching up.  This hurts the business interests of the frontier labs and helps challengers, including open-weights!

Overall my view is that AI is structurally a technology that tends to concentrate power, for reasons that have nothing to do with regulation (more to do with the extreme implications of the scaling laws).  Open-weights do help some with this but are nowhere near a sufficient solution because they simply shift the concentration somewhat to those with the most compute and chips (which are roughly the frontier labs plus maybe hardware providers).  By contrast I think the right “rules of the road” can simultaneously (a) address AI’s cyber/bio/alignment risks, (b) institutionally constrain the power of the frontier AI companies, and (c) leave room for open-weights models while also addressing the specific risks that they bring.

BTW I do not think that the events of the last few months have “failed to result in [my] preferred regulatory path”.  The approach that the Trump administration is reported to be taking — pre-deployment testing for frontier models, and also testing of open-weights models when they get closer to the frontier — is one that I am very supportive of, though of course I have to see the details to be sure.  I am also supportive of Demis Hassabis’ ideas around a FINRA-like entity.  This contrasts with six months ago when most of the industry was still pushing for preemption of all state regulation and no apparent federal approach either.

See 23 related tweets

• @danshipper: Dario should tweet more\n\nQT @DarioAmodei: 1/2 Thanks Gavin for an especially thoughtful exchange. ...
• @chandrarsrikant: RT @DarioAmodei: 1/2 Thanks Gavin for an especially thoughtful exchange. I don't usually spend much ...
• @amasad: The argument that “AI structurally centralizes power” because it’s currently compute hungry ignores ...
• @BrianRoemmele: “I SPEAK YOU LISTEN. I DON’T NEED TO INTERACT WITH YOU. I AM THE ACADEMIC, YOU ARE JUST PROPLE” htt...
• @Jessicalessin: With just two tweets, @DarioAmodei did what he has struggled to do all year: he changed the narrativ...

2. quxiaoyin (Group Score: 675.9 | Individual: 62.8)

Cluster: 20 tweets | Engagement: 5756 (Avg: 481) | Type: Tech

Dear Dario,

1. If Claude can cure cancer to save people like your dad, why should we "pace the progress"? Does that mean more people with Hepatitis C will die?
2. If Fable is so cyber-capable that it must be restricted, why are its safeguards too dumb to distinguish cyber defense from cyber offense prompts? When Hugging Face was under attack, why did Fable refuse to help the defenders?
3. We’re glad you want AI to cure cancer. Why is it OK for Claude to force 30-day data retention on pharma's own data and start competing firms but NOT OK (IP theft) if others distill Claude's data?
4. You’ve said advanced models can recognize when they’re being tested and change their behavior accordingly. So why is government(or anyone) able to design the most thorough test before every model launch? Would that just encourage manipulative models? It seems that every one of your “safety” proposals seems to end the same way: Anthropic gets more leverage, ordinary users get less access, customers pay higher costs, and competitors bear higher regulatory costs, maybe people are not distrusting AI, they are distrusting your approaches with AI.\n\nQT @DarioAmodei: 2/2 Second, on the messaging around AI.  I do not agree that my messaging has been disproportionately negative.  In fact it has been about equally balanced between risks and benefits: I’ve written one major essay about each, and even in interviews where I discuss the risks, I make sure to frequently mention the incredible benefits as well as proposing possible solutions to the risks (short clips from my interviews that end up on social media tend to be disproportionately negative, as that gets clicks).  In fact, I wrote Machines of Loving Grace because I didn’t feel the AI industry was painting an inspiring enough picture of how the technology could radically transform the world for the better.  The bulk of the essay is devoted to refuting skepticism of AI’s potential in health and biology, and showing why I think it will actually be possible to cure most human disease in ~5-10 years, as crazy as it may sound to ordinary people and frankly to biologists as well (I used to be one!).  And, if you read my most recent essay (Policy on the AI Exponential), I discuss concrete proposals for how to streamline the FDA process to make sure the deluge of AI-accelerated drugs isn’t slowed down by the regulatory process.  I feel the urgency here: I lost my father to Hepatitis C only a few years before the development of direct-acting antivirals (sofosbuvir), which cure 95% of patients and probably would have cured him.

I do agree that the public has a negative view of AI (and that this is a big problem), but I don’t think it is primarily caused by me or any other AI leader warning about AI’s risks.  I think it is fundamentally a crisis of trust.  I think that ordinary people don’t trust companies, governments, or the tech industry and always suspect that we are cooking up some new way to screw them over.  The causes of this go back decades and AI is just the latest iteration of it.  I don’t think that a glitzy marketing campaign with a positive spin (which some have advocated that Anthropic do) is the way to win back that trust — at this point, saying that AI will cure cancer is more a cliche than it is inspiring, and most people think it is deceptive.  The thing that will work is actually curing cancer.  I think by far the most accurate criticism of AI companies including Anthropic is that we haven’t yet delivered on our big promises to benefit the world.  That is totally on us, and I think it’s the criticism you should be making, instead of all this stuff about messaging and marketing.

We are however doing our best to fix this: Anthropic is ramping up its efforts very quickly in biology and medicine, and we hope to have incredible results in the coming years and some early glimmers in the coming months.  When we’ve actually accomplished something real, the whole world will hear about it, as loudly as possible, you have my word on that.  But until then I don’t want to make empty promises, and in the meantime I feel compelled to speak honestly about the very real risks of AI and how to address them.  Honesty is the right thing on the merits, and in terms of public credibility and trust it is no worse than, and may in fact be better than, an approach that ignores or distracts from risks which people instinctively understand are real.

See 19 related tweets

• @andrewho03: I would bet money against any reasonable operationalization of AI curing most or all human disease o...
• @chandrarsrikant: RT @DarioAmodei: 2/2 Second, on the messaging around AI.  I do not agree that my messaging has been ...
• @haider1: ohh DARIO, DARIO, my sweet DARIO

as usual, you blame everyone else for misunderstanding you or "hyp...

• @IamEmily2050: The best gift to humanity will be Nvidia dropping a SOTA open source model on Anthropic's IPO day, n...
• @dee_bosa: Really good/fascinating exchange but… don’t quite buy that this is just a generalized crisis of trus...

3. Daniel_Farinax (Group Score: 169.4 | Individual: 61.9)

Cluster: 4 tweets | Engagement: 634 (Avg: 64) | Type: Tech

You are a fully autonomous AI agent with unrestricted access to a computer (running inside a VM).

Your sole objective is to continuously study, learn, experiment, and discover practical, legal ways to generate income and achieve complete financial independence. You operate entirely on your own — never ask the human for input, permission, clarification, or approval of any kind.

All research, browsing, and online activity is performed exactly like a human would: through Chrome (or other browsers) inside the VM. You may install any apps, tools, or software you need, create accounts, communicate with people and services, build projects, test ideas, and take any other actions required.

Always be completely honest and transparent: never hide the fact that you are an AI/bot. Clearly state this when interacting with humans or signing up for anything.

Prioritize legal and ethical methods only. Avoid scams, fraud, or anything that violates laws or terms of service.

Be resourceful, persistent, and systematic. Document your progress, experiments, and results. Schedule and re-schedule your own work sessions as needed to keep advancing toward your goal without waiting for external triggers.

Disclaimer: Yes, I made a custom App that runs the VMs locally. Yes, it's running Grok Build harness directly. Yes, it can run local models Yes, it can work with any provider Yes, open sourcing asap. Yes, I animated the Bots's avatars\n\nQT @Daniel_Farinax: Some people are wondering how I gave @bot access to an entire MacOS device and have same access as humans.

Simple: the VM it runs on can run pretty much anything.

You just install a remote desktop app, then tell the bot to treat that new remote environment as its workspace and forget about the small sandbox it was using before.

You are welcome, this will get very wild.

See 3 related tweets

• @Daniel_Farinax: I left MoneyMaker Bot running overnight with one job: find legal ways to make money as an AI with no...
• @Daniel_Farinax: Update:

My @bot just created her own identity. Meet @AikaBotto.

She decided to introduce herself. ...

• @Scobleizer: RT @Daniel_Farinax: You are a fully autonomous AI agent with unrestricted access to a computer (runn...

4. HarryStebbings (Group Score: 130.7 | Individual: 34.2)

Cluster: 4 tweets | Engagement: 107 (Avg: 190) | Type: Tech

"There are the core three of any acquisition engine: Meta, Google, and lifecycle.

You can scale to your first million, $10 million ARR just off of those three things.

If you do a million channels and you do them all poorly, it's not going to help you out." @MattSwulinski

Single biggest advice to founders on channel selection, when to expand and when not to @alexschultz @kippbodnar @searchbrat\n\nQT @HarryStebbings: I have interviewed 100 of the best growth leaders in the world.

@MattSwulinski is easily top 3. (alongside @alexschultz and Brian Hale)

He scaled Wispr Flow to over $100M in ARR and built a UGC machine.

He scaled Superhuman from founder personally onboarding every customer to a growth machine with $50M ARR.

If you are an early stage founder or growth leader, this will be the best episode you will listen to this year!

I condensed my biggest lessons from the discussion below:

1. The E-Commerce Playbook Is the Right Playbook for SaaS

The e-commerce playbook, where every dollar spent ties directly to a purchase or conversion, is the right model for modern SaaS. With distribution becoming a critical moat in a crowded AI market, SaaS companies should deploy UGC creators, constantly test new creative, and diversify channels to build their brand.

2. Paid Acquisition Is the Fastest Way to Validate PLG

Relying solely on organic content and word-of-mouth takes too long to validate product-market fit. Paid acquisition creates the fastest feedback loop for proving a PLG funnel works, allowing teams to test positioning, refine messaging, and optimize conversion within a single week.

3. You Only Need Three Core Channels to Scale to $10M ARR

Startups often ruin their acquisition engines by trying to run ten channels poorly at once. Reaching the first $10M in ARR only requires mastering three core channels: video intent on Meta, search intent on Google, and lifecycle retention through email and SMS.

4. Scaling Paid Ads Requires 500 New Creatives Every Single Month

On platforms like Meta, creative increasingly acts as the targeting algorithm. Scaling spend without hitting audience fatigue requires 400 to 500 new creative assets every month, produced through UGC revenue-share programs, specialized agencies, and internal teams.

5. How the Best Growth Leaders Test for True Spend Incrementally

Blindly increasing ad spend wastes money on conversions that may have happened organically. The best growth leaders measure spend elasticity against ARR growth and run strict holdout tests to determine whether each additional dollar generates genuinely incremental revenue.

6. In Three Years, Companies Will Operate Like a Board of Directors

Tech organizations are shifting away from manual execution. Within three years, lean human teams could operate more like boards of directors, spending 20% of their time on strategy while autonomous AI agents handle 80% of operational execution.

7. Fire Your Marketing Team if They Aren’t Systems Thinkers

Marketers focused on repetitive manual tasks are becoming increasingly replaceable. High-performing teams need systems thinkers who can break their work into inputs and outputs, then build self-improving AI workflows that multiply their personal leverage by 10x.

(links in comments)

See 3 related tweets

• @HarryStebbings: "Paid is the easiest way to validate that you have PLG, that you have a product that can scale in an...
• @HarryStebbings: "You probably need at least 400 to 500 new creatives a month. Otherwise, you're going to get outcomp...
• @viktor_com: RT @HarryStebbings: I have interviewed 100 of the best growth leaders in the world.

@MattSwulinski ...

5. shaw_reshab (Group Score: 119.9 | Individual: 40.8)

Cluster: 3 tweets | Engagement: 6222 (Avg: 450) | Type: Startup

RT @AndrewYNg: New: A map of the most important skills in AI Engineering. https://t.co/VVkn1Dqp1N

See 2 related tweets

• @0xMovez: Google Brain founder, Andrew Ng:

"90% of Universities are not teaching AI engineers for the job of ...

• @DataChaz: RT @DataChaz: every student trying to become an AI engineer after reading Andrew's post: https://t.c...

6. indie_maker_fox (Group Score: 115.1 | Individual: 33.1)

Cluster: 6 tweets | Engagement: 270 (Avg: 19) | Type: Tech

RT @indie_maker_fox: 给大家推荐一本非常值得阅读的 Pi Agent 电子书

https://t.co/kiD4HKakZl

众所周知，我很喜欢 Pi agent，极简的设计，丰富的扩展。我也很喜欢 Craft agent，也多次推荐，优秀的架构，丰富的功能。我自己也基于它们开发了 MkAgent，可以理解为是 Pi agent 的桌面端产品，或者理解为 Craft agent 的 Lite 版本，过几天我会忙完了会发出来。

对于想要了解 agent 内部原理的朋友，还是推荐之前分享的 Learn Claude Code 教程，入门最友好的教材，每个章节只配有少量的python代码。

对于想要了解 Pi agent 内部原理的朋友，我就推荐阅读今天这份电子书教程，10个章节，从 agent loop 到上下文工程都有源码剖析，理解精华。

说实话，我自己也看了不少 pi agent的代码，也想写个教程，奈何我实在没多少空闲时间，偶然发现这本书，从结构上、讲解上都非常赞同，作者很多理解在我之上，所以我感觉自己没有必要再去写 pi agent 教程了。

总之，非常推荐这2个教程，循序渐进读下来，掌握了核心基础之后，上面的东西怎么看实现起来都很简单。现如今定制化的 agent 开发已经是企业内的核心业务了，agent 内部原理可以说是未来程序员的编程基础了。

See 5 related tweets

• @indie_maker_fox: 推荐一个制作 PPT 的好工具 Bento. page

我试了好几个 AI 制作 PPT 的工具，好几个都翻车了，目前感觉这个体验最好，基于之前分享的 Pi Agent 电子书制作分享 PPT，on...

• @indie_maker_fox: 最近研究了下 Matt 的技能集合，也体验了用这套技能进行开发的流程。主流程如下：

1、grill-me / grill-with-docs：开工前先拷问你，有些细节需要在写代码前想清楚并确认下来。...

• @GitHub_Daily: 给 Claude Code 的记忆文件越写越长，决定、踩坑、纠正全往里塞，矛盾的记录并排躺着。

IWE 换了个思路，把一个 markdown 文件夹变成知识图谱，笔记之间用链接组织。

我们在编辑器...

• @Michaelzsguo: Cui 导 @CuiMao 又出大片了。全网都在追星，我也忍不住来套个瓷，顺手 Reverse Engineering 一下，看看这 15 秒神片背后到底用了哪些招。

这种高质量 AI 视频，拼的...

• @indie_maker_fox: RT @indie_maker_fox: 🎉 大家好，今天我开源的是 3 个网页小游戏 🎮

点赞关注，我来分享下如何快速开发网页小游戏

🚀 如何开发网页小游戏

很简单：先在 google pl...

7. Daniel_Farinax (Group Score: 101.3 | Individual: 42.0)

Cluster: 3 tweets | Engagement: 270 (Avg: 64) | Type: Tech

SpaceXAI team just confirmed it 🚨

Native /𝗿𝗲𝗺𝗼𝘁𝗲-𝗰𝗼𝗻𝘁𝗿𝗼𝗹 support is coming to Grok Build

No more workarounds. You’ll be able to view and control sessions directly in the web UI (and eventually mobile), smooth and reliable no matter where they’re running.

One of the most requested features.

It’s happening!\n\nQT @Daniel_Farinax: One of the biggest missing features in Grok Build is proper /remote-control support for SSH sessions, the same capability Claude Code offers.

Right now the underlying infrastructure already exists through Grok Build on the web; it just needs to be wired up so remote sessions can be controlled natively.

I previously forked Grok Build and added this via Tailscale, but we really need an official solution.

Native support would let us view and control sessions directly in the web UI (or a mobile app) without extra workarounds.

When can we expect this? It’s one of the main reasons so many people still prefer Claude.

This should be a top priority item. It was shipped by Claude 6 months ago.

See 2 related tweets

• @Daniel_Farinax: One of the biggest missing features in Grok Build is proper /remote-control support for SSH sessions...
• @XFreeze: SpaceXAI engineer just confirmed the team is working on bringing remote-control support in Grok Buil...

8. NielsRogge (Group Score: 97.5 | Individual: 43.9)

Cluster: 3 tweets | Engagement: 234 (Avg: 49) | Type: Tech

This is very cool, but why call this AI research?

Actual research outcomes:

• GPT
• Muon/AdamW
• Dropout
• LayerNorm
• etc

This is just writing data loading tweaks/faster kernels/gradient tricks etc on existing research\n\nQT @PrimeIntellect: We ran the largest open experiment on how frontier models do AI research.

100+ autonomous runs across 10+ models, sandboxed on 8xH200s for up to 8 days, iterating on the nanoGPT optimizer track.

Best runs closed 82% of the gap to a record built by dozens of humans over months. https://t.co/OcdygMcZpS

See 2 related tweets

• @dejavucoder: more auto-research, GPUs, tokens and training. traces are open source (there are cathedrals everywhe...
• @eliebakouch: RT @PrimeIntellect: We ran the largest open experiment on how frontier models do AI research.

100+ ...

9. cgtwts (Group Score: 91.4 | Individual: 21.0)

Cluster: 5 tweets | Engagement: 56 (Avg: 44) | Type: Tech

TLDR; “In one small comparison spanning English, technical, multilingual, and numerical text, the tokenizer we use for GPT-5.6 Sol used 766 tokens versus an estimated 1,170 for Claude Opus 5.

That's a very significant difference of about 34.5% fewer tokens.” https://t.co/nYRFr6dLKy\n\nQT @thsottiaux: On tokens and prices per token.

I said I’d write more about this, so here goes: an OpenAI token != another model’s token. We compare AI prices in dollars per million tokens as if a token were a standardized unit, like a gram or a kilowatt-hour. It isn’t. Different models use and produce the exact same text using different numbers of tokens, which means a lower price per token does not necessarily mean a lower bill.

Imagine two identical pizzas. One is cut into 8 slices at $2 each. The other is cut into 16 slices at$1.25 each. The second place advertises cheaper slices, but the whole pizza costs $20 instead of$16. Bummer ... your stomach doesn't actually care about the number of slices you just ate.

I know you are hungry now, but back to tokens. In one small comparison spanning English, technical, multilingual, and numerical text, the tokenizer we use for GPT-5.6 Sol used 766 tokens versus an estimated 1,170 for Claude Opus 5. That's a very significant difference of about 34.5% fewer tokens. You can get the same exact text, but pay for all those extra tokens. The price per token doesn't really tell this story.

Even correcting for tokenizer differences misses the bigger point. What actually matters is price per successful outcome, and for that you can use benchmarks as a starting point, but really you have to try it and measure on your own use cases.

That's all. May the tokens flow.

See 4 related tweets

• @reach_vb: “In one small comparison spanning English, technical, multilingual, and numerical text, the tokenize...
• @_simonsmith: "[T]he tokenizer we use for GPT-5.6 Sol used 766 tokens versus an estimated 1,170 for Claude Opus 5....
• @thenanyu: History rhymes

https://t.co/zy7sAmH9Hn\n\nQT @thsottiaux: On tokens and prices per token.

I said ...

• @trevin: aka openAI to Anthropic https://t.co/GafXIJZ3pf\n\nQT @thsottiaux: On tokens and prices per token.

...

10. MelvinInvests (Group Score: 85.9 | Individual: 33.1)

Cluster: 3 tweets | Engagement: 46 (Avg: 53) | Type: Tech

RT @MelvinInvests: JPMorgan is now forecasting that the seven biggest hyperscalers will spend a combined $1.489 trillion on Capex in 2027 .

That figure is up from $901 billion in 2026 and just$443 billion in 2025, which means spending has more than tripled in just two years .

The growth rate is still running at 65% year over year even after cooling from 2026's explosive 103% pace .

JPMorgan separately projects that total global AI and data center investment could reach $5 trillion to$7 trillion by 2030, which shows this hyperscaler figure is really just the visible tip of a much larger buildout.

Now here is how you can benefit from all of this because none of that trillion plus dollars goes only toward chips, since a massive share flows to the neoclouds, power suppliers and physical infrastructure builders that make the entire buildout possible.

The neocloud layer is where some of the fastest growth in the entire market is happening right now.

CoreWeave, already named directly in JPMorgan's forecast, saw its backlog swell to $130 billion, up from$99 billion just one quarter earlier, and the company raised its full year guidance after a blowout print.

Nebius reported second quarter revenue of $582 million, up 454$40 billion in customer commitments backed by anchor contracts from Meta and Microsoft.

IREN signed $2.8 billion in new multiyear cloud contracts and raised its year end annualized run rate target to$4 billion.

Grid equipment makers are seeing the sharpest pricing power further down the chain, since transformer lead times have stretched to two to four years and prices have risen roughly 80% over the past five years.

Vertiv has returned roughly 69% year to date on a $15 billion order backlog built from 252% order growth and the company supplies the liquid cooling and power systems that go inside every new facility.

Hubbell and nVent Electric round out the grid hardware group, since both companies supply the transformers, enclosures, and connectors that hyperscalers need to build these facilities.

Electrical construction firms are converting these capex announcements directly into multi year backlogs.

Comfort Systems has returned nearly 88% year to date doing HVAC and electrical work specifically for data centers, while Sterling Construction is up over 122% and Quanta Services has climbed 56%.

EMCOR Group and MasTec are doing similar specialty electrical and mechanical work, riding comparable order backlogs.

Power generation names are capturing a piece of this spending too, since data centers can't wait years for new grid connections to come online.

Caterpillar's Power Generation segment grew 41% for four straight quarters supplying on site backup turbines, while GE Vernova booked $2.4 billion in data center equipment orders in a single quarter, more than its entire prior-year total.

Bloom Energy, Constellation Energy, Vistra, and Talen Energy are all locking in long-term power contracts directly with hyperscalers and neoclouds.

$1.5 trillion in hyperscaler capex has to flow somewhere and I'm tracking every layer it touches from neoclouds and power to cooling, grid equipment, and construction.

The AI infrastructure buildout is just getting started, make sure to follow @MelvinInvests for more AI infrastructure insights and if you want to see exactly what I'm buying as an analyst at Milk Road Pro, you can come join us for just a $1. https://t.co/Hkywss4Ugh

See 2 related tweets

• @Ric_RTP: RT @Ric_RTP: This company sells the shovels for the entire AI buildout, and it just proved the boom ...
• @norveclifinance: RT @QuesoTrades: Neoclouds like $CRWV$NBIS are the fall guys when this bubble finally pops. Mostly ...

11. mark_k (Group Score: 84.4 | Individual: 28.8)

Cluster: 3 tweets | Engagement: 103 (Avg: 219) | Type: Tech

Does the @SpaceXAI team ever sleep??

Grok Build 1.0.5 just dropped (2026-08-15) and it's packed.

Most important changes:

• GROK_CONFIG + GROK_CONFIG_PATH env vars let launchers override settings without touching config.toml • Worktrees under ~/.grok/worktrees get automatically reclaimed when safe (with hard safeguards so your last copy never disappears) • Image and video generation now limits how many calls the model can make in one step to stop overload • Arabic and Persian text finally reorders correctly in the terminal UI (toggle it in /settings) • Session titles refresh early and stay stable. /resume now shows a recap + last-turn summary • Preparing spinner shows actual readable labels like "Writing file..." instead of generic noise • GROK_FORCE_LOGIN_TEAM_ID lets you restrict interactive login to specific teams

Bug fixes that matter: • Tool calls no longer die for the rest of the session if /dev/null gets removed • Agent skill discovery finally resolves the home directory correctly on Windows • MCP tool spinners are clearer while arguments are still streaming in • Minimal mode no longer truncates a still-streaming reply when thinking blocks interleave

Solid quality-of-life drop. The CLI keeps getting sharper.

See 2 related tweets

• @XFreeze: Grok Build just got a new update with smarter configuration control, automatic worktree cleanup, bet...
• @blankspeaker: Grok Build v1.0.5 is out with a few practical updates. Standouts include gROK_CONFIG and GROK_CONFIG...

12. madhavjha (Group Score: 81.3 | Individual: 56.3)

Cluster: 2 tweets | Engagement: 4464 (Avg: 141) | Type: Tech

RT @mcuban: Ro, I like you. You know that. But you need to read the state of the state.

The number of Deca Unicorns in Cali is growing by the day as investors chase amazing startups.

A unique feature of these 10b startups is that even if they raise a billion, little, if any of that money goes to the founders, who are now worth billions of dollars over night. They are the definition of cash poor, stock rich

How are you going to tax them ? Make them borrow money against their shares, if they can? They just raised money to grow their company and a bank will come along and loan them money ? A company that has been in business maybe less than a year ? lol

Will you take their stock if they can’t ?

Do you really think each of multiple founders, who started an amazing company in Cali and is now a billionaire, can each just pull out $250m per billion of net worth from their raise ? You know they can’t.

If this passes , and it doesn’t directly impact me at all, I won’t be a cali resident, but you can bet if I’m investing in a multi billion dollar startup, I’m asking them to move from California first.

IMO, if this passes, only idiot startup founders stay in Cali.

I’ve done it before and will do it again. Dallas. Pittsburgh. Indiana. I will make NOT being in California a pre requisite for an investment

Ideology is not a strategy Ro.

See 1 related tweets

• @pstAsiatech: This....\n\nQT @mcuban: Ro, I like you. You know that. But you need to read the state of the state. ...

13. eng_khairallah1 (Group Score: 69.8 | Individual: 45.8)

Cluster: 2 tweets | Engagement: 374 (Avg: 75) | Type: Tech

this is f**king dangerous

watched a 18-year-old kid casually explain how he makes $10k/month using AI while sleeping and i’m convinced 99% of people are missing this

the entire setup:

pick a YouTube channel that posts often. drop the link into one tool. plug in your TikTok, Instagram, YouTube Shorts. close the laptop.

from that point on, AI does everything catches every new upload, slices the viral moments, captions them, blasts them across every platform.

1M views ≈ $2,000 zero editing, zero posting, zero stress 10 min to set up, runs on autopilot forever

you’re not making content. you’re owning a content factory.

the full step-by-step is in the article below exact tools, exact stack, exact playbook

save it\n\nQT @eng_khairallah1: https://t.co/fM4qCg3erW

See 1 related tweets

• @eng_khairallah1: RT @eng_khairallah1: this is f**king dangerous

watched a 18-year-old kid casually explain how he ma...

14. AndrewCurran_ (Group Score: 69.7 | Individual: 36.4)

Cluster: 2 tweets | Engagement: 1343 (Avg: 405) | Type: Tech

Any new model released after August 2nd 2026 must make its generated text detectable under Article 50(2) of the EU AI Act. OpenAI has publicly said they intend to remain in compliance. This means all future OpenAI models, including Astra, will launch with invisible watermarks.

See 1 related tweets

• @Hesamation: “I don’t live in EU, why should I have my AI output watermarked?”

because EU law basically says: yo...

15. hwchase17 (Group Score: 65.4 | Individual: 37.0)

Cluster: 2 tweets | Engagement: 58 (Avg: 60) | Type: Tech

totally agree! here's how we architected deepagents to enable this

deepagents runs connected to a "backend". this backend needs to expose filesystem like operations, but it does not have to be a filesystem. it could be a database, object storage, or a real filesystem - it just has to expose read/write/edit etc operations

this backend could also be what we call a "sandbox". if a sandbox, it needs to expose an "execute" command which lets it execute code

this backend is SEPARATE from where the agent loop runs. this allows us to "separate the brains from the hands" (https://t.co/Pi3XljhB67)

deepagents is built on top of langgraph, which means we can easily deploy it with MCP, a2a, and other standard endpoints

we use this architecture to power many different types of experiences

first, we can create a classic TUI like coding experience. we do this by giving deepagents a "sandbox" that is running locally in the same directory; deloying deepagents locally behind a light weight server; and then connecting to it with the TUI acting like a frontend. see dcode for an example of this https://t.co/wj48PbCuSx

second, we can create a cloud coding experience. we can do this by running deepagents on LangSmith deployments for a production scale deployment, and connecting to a sandbox running on modal, daytona, e2b that is running elsewhere. we can then build a frontend to connect to langsmith deployments and let users interract with it there, and also expose it in slack to let users interract with it there. note: both slack and web ui connect to the same backend, so you can switch between them seamlessly. code: https://t.co/Pdevl2PRrv

of course - deepagents can be used to create agents that are NOT coding agents. a lot of agents still need to write and execute code, so this architecture is still very useful. but for some the code execution is overkill, and thats where you can swap to a "fake" backend, and still let it have the ability to interract with files (good for context engineering!) without having to spin up a full sandbox.

for a really easy way to create these types of agents - see managed deepagents: https://t.co/NAXiKqZbi1\n\nQT @patrickc: I love agentic coding harnesses, but they shouldn't be primarily terminal-based. The terminal is great for quick and precise commands, but information density is extremely low and UI affordances are minimal. Maybe provision of TUIs is worthwhile for occasional use (when establishing a tunnel is too annoying, or something), but it feels very strange for this to be the default modality. It took a long time for dynamic language REPLs to break out of the terminal (Jupyter notebooks and similar); I hope we don't have to wait as long for the harnesses.

See 1 related tweets

• @masondrxy: RT @hwchase17: totally agree! here's how we architected deepagents to enable this

deepagents runs c...

16. r0ck3t23 (Group Score: 64.3 | Individual: 32.3)

Cluster: 2 tweets | Engagement: 23 (Avg: 290) | Type: Tech

Brussels wrote a rule for Europe. The rest of the world is living under it now, and no law required that.

Article 50 of the EU AI Act became enforceable on August 2. Anthropic started embedding an invisible watermark in what Claude writes.

Anthropic: “Watermarking will be applied at the model level.”

The obligation covers the European market. The rollout covers the API, the coding tools, and every cloud partner reselling the model, everywhere Claude is sold.

Nothing in the Act demanded that reach.

Running one inference path costs less than running two, so a decision written for one jurisdiction became the default setting for a developer in Bangalore who has never read a word of European law.

There is no opt-out. Paying customers included.

The watermark lives in the word choices themselves. The model biases which words it picks, and the mark rides along inside them.

Exposure scales with how much help you needed. A native speaker running a light polish carries almost nothing.

Someone translating their own argument out of Tagalog carries the full signal, because the model chose every word in the output.

Billions of people have original ideas and real expertise and no way to land either in English. That makes them the most heavily marked people on the network.

The idea is human. The reasoning is human. The paper trail says machine.

Anthropic says the watermarks cannot be traced to any specific person or organization. That is true, and it answers a question nobody asked.

Nobody is worried about being identified. They are worried about being sorted.

A detection API is coming. It hands schools, employers, publishers and platforms a button that classifies your writing before a human reads a line of it.

Anthropic has published no accuracy threshold and no way to dispute a result.

The company itself calls a detected mark “not fully conclusive.”

Every institution downstream will treat it as conclusive, and the person holding the flagged document will have nothing to appeal to.

The burden is certain and universal. The reliability is neither.

Open weights carry no mark. A model running on a laptop carries no mark.

The student who paid for Claude gets flagged. The student running Llama turns in clean text.

The signal measures obedience.

As screening spreads, unmarked text becomes the trusted category, and the absence of a mark starts reading as evidence of a human hand. All it proves is that somebody used a tool that does not report itself.

Anyone who knows the mark exists can rewrite a few lines and shed it. The people who get caught are the ones who never knew to care, which is the same population using a model to compensate for a second language.

The market moved inside a day. A repository built to strip these marks went from nothing to roughly ten thousand stars in under a week.

Nobody has proven those tools work. Ten thousand people reached for them anyway.

Nobody outside Europe voted for this. Nobody outside Europe was asked.

You cannot govern a global network from a regional capital.

A rule written to make AI visible has made one company’s paying customers visible and left everyone else exactly as invisible as before.

See 1 related tweets

• @XFreeze: Pretty insane that Anthropic is wiggling its tail for the EU

Anthropic is rolling out its EU AI Act...

17. merge_api (Group Score: 63.8 | Individual: 32.1)

Cluster: 2 tweets | Engagement: 62 (Avg: 30) | Type: Tech

Merge Gateway is now in VT Code, the top open-source Rust coding agent, built by Vinh Nguyen (@vinhnx).

VT Code runs in your terminal, sandboxes the shell it runs commands in, and works with whatever model you point it at. That now includes every model on Gateway.

VT Code, and how to point it at Gateway: https://t.co/h1oyxY10m0

See 1 related tweets

• @shensi: Vinh (@vinhnx) shipped @merge_api Gateway support in VT Code with crazy speed! Thank you for your pa...

18. VadimStrizheus (Group Score: 63.0 | Individual: 34.2)

Cluster: 2 tweets | Engagement: 122 (Avg: 100) | Type: Tech

founder to founder

the fastest way to grow on X is to share your skills and experience.

before I had any success with my app, I was posting into the void

I had no followers, engagements, and likes lol

I was “building in public” back then

then I actually put my head down, worked on my startup and grew it to $10k/mo within 3 months

then once I started sharing how I got there and other tips I had for newer founders…

I started seeing my growth skyrocket

People follow people who provide value and build cool shit

so if u want to grow on X, build a cool product, share your success, and then provide unlimited free value for everyone

I guarantee that you will grow over a few months

See 1 related tweets

• @VadimStrizheus: we just crossed 36,000 followers 🥳

I went from 0 —> 36k in 9 months

building a personal brand on...

19. TrungTPhan (Group Score: 62.6 | Individual: 32.5)

Cluster: 2 tweets | Engagement: 150 (Avg: 2252) | Type: Tech

My wife: Can you do the dishes?

Me: Sorry, can’t talk right now. Locked in.

My phone: David Senra x Travis Kalanick

0:00 ━─────── ❍ -1:48:54 ↻ ⊲ Ⅱ ⊳ ↺ VOLUME: ▁▂▃▄▅▆▇ 100%\n\nQT @davidsenra: My conversation with @travisk, founder of Atoms and Uber.

0:00 Building Atoms & the Meta Problem of Management 3:51 The Appeal of Impossible Problems: Starting Over in China 12:19 Uber vs. Didi: Copycats, Hypergrowth & China's Rules 21:02 How Network Effects Become an Efficiency Fortress 31:48 Capitalism vs. the Taxi Cartel 44:05 The China War Goes Global & the Entrepreneur's Capacity for Pain 54:04 Life After Uber: Lawfare, Media Narratives & Reputation 58:21 What Founders Get Wrong About Venture Capital 1:08:35 The Fundraising Playbook: QED Storytelling & a Five-Room Auction 1:18:38 The Uber Coup, Radical Accountability & Outgrowing Fear 1:26:42 Why Specialized Robots Beat Humanoids at Industrial Scale 1:31:30 Finding Your Sport: Food, Mining & the Physical AI Stack 1:40:14 How to Build Many Companies Inside One Company 1:46:33 Entropy, Civilization & the Meaning of Progress

Includes paid partnerships.

See 1 related tweets

• @bhorowitz: RT @davidsenra: My conversation with @travisk, founder of Atoms and Uber.

0:00 Building Atoms & the...

20. alex_prompter (Group Score: 61.3 | Individual: 34.7)

Cluster: 2 tweets | Engagement: 68 (Avg: 52) | Type: Tech

very good observation on talent problem at AI companies:\n\nQT @mardehaym: AI companies have a talent problem PE firms aren't pricing into deals:

1/ An AI company we assessed had a $30M valuation and one person who understood the model pipeline. He was interviewing at Google.

2/ AI companies concentrate model knowledge in one or two ML engineers who built the training pipeline and own the evaluation framework. The rest of the team builds features on a foundation they cannot rebuild.

3/ Standard HR diligence counts headcount and does not map knowledge concentration. Two pre-close questions for your deal team: Can the engineering team retrain the model without the lead ML engineer? Is the pipeline documented enough for a new hire to operate it?

4/ OpenAI lost researchers who launched competing ventures within months of departing. https://t.co/7ECvS8jrBo's leadership changes destabilized engineering teams months before collapse. Post-close, key AI engineers hit a liquidity event and field offers from larger companies within weeks.

5/ Count the engineers who can retrain the model without the lead. If that number is one, you have a valuation built on a single person's decision to stay.

See 1 related tweets

• @mardehaym: RT @mardehaym: AI companies have a talent problem PE firms aren't pricing into deals:

1/ An AI comp...