科技推特精选 - 2026年5月11日

2026年5月11日 科技每日简报

Today's top tech conversations are led by @richardczl, whose post about 'RT @eladgil: People at major A...' garnered the highest engagement. Key themes trending across the top stories include center, https, power, there, people. The community is actively discussing recent developments in AI, engineering practices, and startup strategies.

1. richardczl (Group Score: 123.5 | Individual: 40.5)

Cluster: 5 tweets | Engagement: 796 (Avg: 109) | Type: Tech

RT @eladgil: People at major AI labs (using internal models) 3-4 months ahead of startup silicon valley engineers

SV founders/eng 3-6 months ahead of NY

NY founders/eng 6-12 months ahead of rest of world

Most people have no idea how fast AI shifting as 1-2 years behind SOTA

"The future is here, just not equally distributed" - Robert Heinlein

See 4 related tweets

• @OpenRouter: Try out Fusion if you want to get closer to the actual frontier: https://t.co/NVCSXMIUKm https://t.c...
• @cgtwts: LinkedIn users realizing they’re 12 months behind everyone else in the AI world https://t.co/CzamUrP...
• @Hesamation: also X is 6-12 months ahead of LinkedIn.\n\nQT @eladgil: People at major AI labs (using internal mod...
• @Shashikant86: This guy probably is in deep inside a US bubble 🫧 and and might not heard of China 🇨🇳 and London..\n...

2. sairahul1 (Group Score: 112.2 | Individual: 35.1)

Cluster: 4 tweets | Engagement: 25 (Avg: 19) | Type: Tech

Jensen Huang, CEO of Nvidia:

"People are teaching their agents to fully run a business and make money."

Support. Sales. Operations. Coding. Marketing.

Agents are taking every seat at the table.

Entire companies are being compressed into workflows.

But 99% of agents still forget everything the moment a session ends.

This guy wrote the exact playbook to build AI agents that actually remember ↓

Don't scroll past this.\n\nQT @sairahul1: https://t.co/Q8xjLqwT6t

See 3 related tweets

• @sairahul1: Jensen Huang, CEO of Nvidia:

"Every engineer is going to have and manage hundreds of agents."

The ...

• @sairahul1: RT @sairahul1: Jensen Huang, CEO of Nvidia:

"People are teaching their agents to fully run a busine...

• @sairahul1: RT @sairahul1: Jensen Huang, CEO of Nvidia:

"Every engineer is going to have and manage hundreds of...

3. firstadopter (Group Score: 95.5 | Individual: 35.5)

Cluster: 4 tweets | Engagement: 442 (Avg: 61) | Type: Tech

RT @LipBuTan1: Congratulations to my good friend Jensen Huang on being awarded an Honorary Doctorate in Science and Technology from @CarnegieMellon University for his outstanding contributions to accelerated computing and Artificial Intelligence. It was my honor to place upon him his doctoral hood this morning. @intel and @nvidia are collaborating to develop exciting new products!

See 3 related tweets

• @firstadopter: "@intel and @nvidia are collaborating to develop exciting new products!" $INTC$NVDA\n\nQT @LipBuTan...
• @nvidianewsroom: RT @CarnegieMellon: This morning, Jensen Huang, Founder and CEO of @NVIDIA, met with Carnegie Mellon...
• @PatrickMoorhead: This is good.\n\nQT @LipBuTan1: Congratulations to my good friend Jensen Huang on being awarded an H...

4. maks6361 (Group Score: 84.0 | Individual: 46.9)

Cluster: 2 tweets | Engagement: 522 (Avg: 114) | Type: Tech

ASO is dead! Apple has ruined it with ASA updates! There're too many vibecoded apps so can't rank organically! Android users don't pay so android apps suck!

While everyone keeps speaking about that, my wife with help from Claude skills and my support kept building small simple apps, spending 3-4 hours a day.

It took her only 1.5 months to get her MRR from 100$to 1k$.

It's 9 apps in total, revenue is split 50/50 between iOS and Android apps.

No ads, no marketing, nothing. We decided she'll keep shipping apps until she gets to 2-3k$ so we have some budget and then we'll pick a couple of interesting apps and will be growing them with either ads or TT/IG.

So, stop overthinking and asking yourself "is it the right time to start or am I late already". The best time to start was yesterday, but it's still good to start today.\n\nQT @maks6361: My younger son is nearly 2 years old, and my wife’s maternity period is almost over.

So instead of going back to her 9–5, I suggested she become a mobile app developer 😀 so she can work from home and spend more time with us.

She’s very much into design but not a tech person, so I created a few Claude skills for mobile development to help her ship apps.

1 months ago we started the experiment and today she hit her first milestone 🚀🚀🚀

See 1 related tweets

• @seraleev: Not sure if @maks6361 wife has an X account, but together they’re doing something incredible. Their ...

5. BrianRoemmele (Group Score: 69.5 | Individual: 18.0)

Cluster: 4 tweets | Engagement: 73 (Avg: 606) | Type: Tech

THE TECHNOCRAT'S HUBRIS: Not everything can be calculated

THE REALITY CHECK: The belief that narrow optimization can permanently eradicate the flawed human element is the ultimate trap of the Abundance Era.

Machines do not replace the player; they replace the club.

It’s simple https://t.co/vbriGcqteH\n\nQT @BrianRoemmele: As perfect AI and Robots step onto humanity’s most sacred stages to declare "open warfare," we must decide whether to compete on their terms or transcend them. A 1957 radio show proves the "human element” flawed yet resilient can out maneuver the robots.

https://t.co/UbDgjlnUiH

See 3 related tweets

• @BrianRoemmele: Finding the AI and Robot Future Sweet Spot

🚫 Fearful Rejection 🚫 Naive Utopianism ☑️ Pragmatic Inte...

• @BrianRoemmele: In 1957, before there even was the concept of a personal computer let alone personal AI and Robots, ...
• @BrianRoemmele: As perfect AI and Robots step onto humanity’s most sacred stages to declare "open warfare," we must ...

6. Teknium (Group Score: 66.6 | Individual: 35.1)

Cluster: 2 tweets | Engagement: 230 (Avg: 121) | Type: Tech

You can now access the Pareto Code feature from OpenRouter in Hermes Agent, see the docs on use here:

https://t.co/H13fIEDkrn\n\nQT @OpenRouter: Introducing Pareto Code: a new, free, experimental coding router

Set min_coding_score in your request and route to the cheapest code-capable model that clears your bar, ranked by @ArtificialAnlys.

See the Pareto frontier shifting in real time👇 https://t.co/26e1HeGOpp

See 1 related tweets

• @OpenRouter: Docs from @NousResearch on how to set up Pareto Code in Hermes: https://t.co/SEKNXf0sL0 https://t.co...

7. aakashgupta (Group Score: 65.5 | Individual: 34.5)

Cluster: 2 tweets | Engagement: 167 (Avg: 80) | Type: Tech

Kevin O'Leary's new data center is wild.

It will be bigger than San Francisco and draw as much power as all of New York City. 40,000 acres in Utah. 9 gigawatts. $100 billion total spend. Its own natural gas plant burning on-site to feed it.

He says it'll create 2,000 permanent jobs. Critics say data centers create almost none.

Both numbers are real. They don't describe the same building.

A typical hyperscale data center employs 100 to 1,000 people. The Uptime Institute pegs the rate at 1-2 staff per megawatt for automated hyperscale, which puts a 9 GW pure data center at roughly 900-1,800. The 2,000 number is plausible if you count everything inside the fence.

Inside the fence is a lot more than servers.

The on-site gas plant alone runs hundreds of operators. Combined-cycle plants typically need 25-40 staff per gigawatt, which puts 250-360 of the permanent jobs in power generation. Add security, logistics, fiber, and on-site contractor staff, and the 2,000 number stops describing the same job category the critics are pricing.

The approval path is the giveaway. The project went through Utah's Military Installation Development Authority, the same vehicle the legislature created for defense infrastructure. O'Leary's own announcement says target tenants are clients doing work on behalf of national defense, like the US government or tech firm contractors.

A defense-aligned 9 GW industrial campus that leases compute is a different building than a data center.

The framing matters because the tax math hangs on it. Brookings just published the cleanest dataset on this. Hyperscale counties see real information-sector gains, but only when four or more facilities cluster. Wages rise 3-4%. Subsidies in some studies ran over $2 million per permanent job.

A single rural campus with one tenant doesn't get the cluster effect. A defense complex doesn't need to.

Call it what it is. A power plant with servers attached.

See 1 related tweets

• @ShanuMathew93: Going to take the under on them getting anywhere close to 9gw\n\nQT @aakashgupta: Kevin O'Leary's ne...

8. tengyanAI (Group Score: 65.1 | Individual: 32.7)

Cluster: 2 tweets | Engagement: 31 (Avg: 28) | Type: Tech

NoVA (north virginia) is like Manhattan for cloud infrastructure.

everyone wants to be there because everyone important is already there. but if you suddenly need to build a factory the size of a city, Manhattan stops being practical.

so what do you do? you do not abandon Manhattan. you keep the high-value headquarters there and build the huge new capacity somewhere power, land, and permits are easier.

AI training can be more flexible. the GPUs mostly need to talk to each other inside the same cluster

the cluster can be in Texas or Arizona as long as you have cheap, reliable power, land, cooling and fiber. that's what has been happening\n\nQT @tengyanAI: Northern Virginia became the capital of the internet almost by accident!

MAE-East, one of the early hubs of the internet, started in Tysons Corner in 1992. Over time, the center of activity moved to Loudoun. Why?

Fiber attracted networks. Networks attracted clouds. Clouds attracted enterprises. Enterprises attracted more data centers. Add cheap power, tax breaks, federal demand, friendly zoning, and 20 years of local operating talent.

everyone colocated there because everyone else was already there. proximity to your competitors was a product feature. latency mattered esp if customer wanted multi-cloud systems.

But AI is changing the constraint. NoVA (north virginia) was built on fiber gravity. The next AI map is being redrawn around power.

i'm noticing that Loudoun’s substations are tapped. Transmission upgrades take years. Permitting is political. New MW is spilling into Prince William, Stafford, Culpeper, Texas, Arizona, Ohio, Georgia, TVA territory, Iowa, Nebraska, and the Pacific Northwest.

Loudoun is the core of NoVA’s data center empire.

the world’s most important data center cluster has hit its grid ceiling.

(i've been tracking the DC build out and energy on @tessara_ai)

See 1 related tweets

• @tengyanAI: Northern Virginia became the capital of the internet almost by accident!

MAE-East, one of the early...

9. Reuters (Group Score: 61.8 | Individual: 16.7)

Cluster: 4 tweets | Engagement: 71 (Avg: 101) | Type: Tech

Microsoft’s African data center falters on payment demands, Bloomberg News reports https://t.co/AGw4OMWgfn https://t.co/AGw4OMWgfn

See 3 related tweets

• @business: A major Microsoft data center site in East Africa has been delayed by disagreements with the Kenyan ...
• @Polymarket: JUST IN: Microsoft’s planned $1 billion Kenya data center delayed amid dispute over guaranteed gover...
• @FirstSquawk: Microsoft data center in East Africa delayed by disputes with Kenyan government over payment assuran...

10. VaibhavSisinty (Group Score: 57.2 | Individual: 34.1)

Cluster: 2 tweets | Engagement: 51 (Avg: 71) | Type: Tech

While every AI company is fighting over electricity on Earth, Musk just filed the legal papers to put data centers in space.

Welcome to SpaceXAI.

xAI is officially dissolved. All AI is now under SpaceX.

→ FCC application: up to 1 million solar-powered AI satellites.

→ Trademarks filed for "satellite-based data center services."

→ Colossus (220K Nvidia GPUs) already leased to Anthropic the same company Musk called "evil" last year.

→ Grok, Starlink, Starship, AI compute one company. One IPO. $2 trillion target.

Sam Altman called orbital data centers "ridiculous" 2 weeks ago.

This week, Musk turned it into a trademark.

Most people are reading this as a rebrand.

It's actually the most strategic IPO setup in tech history.

PS: Whoever owns space-based AI compute owns the next decade.

See 1 related tweets

• @shiri_shh: bro is selling this domain for cheap.

Elon can easily pay $6M for this or even higher. https://t.co...

11. alexcooldev (Group Score: 56.3 | Individual: 32.2)

Cluster: 2 tweets | Engagement: 100 (Avg: 174) | Type: Tech

Ah I just remembered, hi @jackfriks, I completed your grow app to $5k/month (over$7k/month) challenge at the end of last year 🫡\n\nQT @alexcooldev: My second app just crossed $7k+/month (web + app combined) 🔥. Slow growth every single week, but compounding is real.

Still following the same philosophy: → build 10 apps → make at least $5k/month from each one

(Funny enough, most of them end up doing $10k+/month 😂)

Still ~92% profit margin if I don’t count Apple tax. Probably even higher this month because I stopped buying new phones and just reuse old devices for operations. New phones are only for testing future apps now.

Still an indie hacker. (solo founder) Still no VC money. Still no paid ads.

So profit optimization always matters more to me than looking “big” on the internet 🔥

See 1 related tweets

• @alexcooldev: I’ve said this a lot: my goal is to build 10 apps making $5k/month each, fully organic. No paid ads,...

12. elvissun (Group Score: 56.1 | Individual: 31.8)

Cluster: 2 tweets | Engagement: 28 (Avg: 11) | Type: Tech

if you are not burning 100% of your weekly claude design limit you are leaving money on the table.

takes 5 minutes to:

• lock your design system
• make a new landing page
• polish your onboarding
• make OG image for every page
• animate your launch video
• prettify your SEO pages
• create graphics for social

it's super easy to burn through the entire weekly limit right now

this is probably what will happen in the next 30 days:

1. product matures and the dedicated weekly limit is removed
2. now it burns from your total weekly limit
3. everyone gets $200 in extra usage for the "transition"

enjoy this all you can eat buffet while it lasts\n\nQT @elvissun: claude design is actually insane

made this with one prompt https://t.co/j9ZnMcq20q

See 1 related tweets

• @elvissun: RT @elvissun: if you are not burning 100% of your weekly claude design limit you are leaving money o...

13. zephyr_z9 (Group Score: 54.9 | Individual: 28.4)

Cluster: 2 tweets | Engagement: 181 (Avg: 240) | Type: Tech

"Furthermore, other kernels like the attention indexer & kvcache compressor has been written using TileLang & Triton for fast development velocity."

DeepSeek adopting TieLang will force others to adopt it as well and build an ecosystem around it\n\nQT @SemiAnalysis_: SPEED IS THE MOAT: AMD ROCm software stack has improved performance by over 75x in the last 14 days since DeepSeekv4 launch. The performance comes from fusing mHC operations & also fusing RoPE hadamard transformations to reduce cpu overhead & improve HBM memory utlization. Furthermore, other kernels like the attention indexer & kvcache compressor has been written using TileLang & Triton for fast development velocity.

Another 5x performance improvement is needed to catch up to single node aggregated B200 performance & then another 1.5x is needed to catch up to PD disaggregated B200 performance, which is within the realm of possibility for AMD within the next couple of weeks. Great work to HaiShaw, Thomas, @roaner, @AnushElangovan for this rapid improvement.

See 1 related tweets

• @SemiAnalysis_: SPEED IS THE MOAT: AMD ROCm software stack has improved performance by over 75x in the last 14 days ...

14. MarioNawfal (Group Score: 53.9 | Individual: 21.1)

Cluster: 3 tweets | Engagement: 1274 (Avg: 541) | Type: Tech

Mind-blowing moment from the DeepMind doc: Demis Hassabis gets told AlphaFold can predict all 1-2 billion known proteins in just a month.

He instantly says: “Do it.”

This is how history gets made

https://t.co/TAzjKEWXYP

See 2 related tweets

• @yacineMTB: hero\n\nQT @TrungTPhan: Still incredible that the DeepMind documentary has footage of exact moment ...
• @ShanuMathew93: Coolest part of The Thinking Game\n\nQT @TrungTPhan: Still incredible that the DeepMind documentary ...

15. Prince_Canuma (Group Score: 53.8 | Individual: 29.4)

Cluster: 2 tweets | Engagement: 84 (Avg: 15) | Type: Tech

Feelings aren’t facts!

This is nonsense 😂\n\nQT @MemoryReboot_: Why Mac Studio is a trap for local AI

• Large unified memory looks sexy on paper
• Great for chatbots, terrible for 24/7 tool calling workflows
• No CUDA = no vLLM, no SGLang, no tensor parallel
• $10k plus for a dead end box you can't upgrade

It's like owning a Porsche with a 100 km/h top speed

See 1 related tweets

• @nummanali: Mac Studio isn’t the game

It’s the M3/4/5 in MacBook Pros with 32GB Ram+

With Qwen 3.6 and Gemma ...

16. seraleev (Group Score: 53.8 | Individual: 31.1)

Cluster: 2 tweets | Engagement: 148 (Avg: 52) | Type: Tech

I acquired 3 apps for $62K. The most I’ve ever spent on apps.

Actually did this over a month ago. Been quietly working on updates since: improving some, relaunching others under new brands.

Wine ID had revenue but bad reviews. A basic AI wrapper. I liked the audience and agreed on the price. The plan is to turn it into a Vivino competitor with my own wine database. Big, complex challenge. Exactly the kind I like.

The other two came from a developer who burned out and decided to quit entirely. He just said, take them for whatever.

Private deals. No marketplaces. Developers made me an offer, I said yes.

See 1 related tweets

• @seraleev: POV: buy 3 apps to 10x them in a year https://t.co/DaVoZYOUBB\n\nQT @seraleev: I acquired 3 apps for...

17. 0xSero (Group Score: 53.7 | Individual: 30.1)

Cluster: 2 tweets | Engagement: 63 (Avg: 252) | Type: Tech

I cannot begin to explain the pain I have suffered dealing with all the provider specific insanity.

Every harness, every agent has to deal with this issue. Opencode and Pi are really building open source agents with intentionality.

Thank you\n\nQT @thdxr: we're working on a library to abstract over all the llm providers

there's very few teams that have dealt with the quirks between providers at the scale we have

it's written in effect but will also have a vanilla api

progress is in the opencode repo under packages/llm

See 1 related tweets

• @nummanali: RT @thdxr: we're working on a library to abstract over all the llm providers

there's very few teams...

18. XFreeze (Group Score: 53.4 | Individual: 29.0)

Cluster: 2 tweets | Engagement: 656 (Avg: 947) | Type: Tech

Elon Musk clearly explains why Grok must be maximally truth-seeking:

“I think you want Grok to say things that are true, even if they’re not politically correct”

Focus on cogent reasoning - make sure axioms as close to true as possible, no contradictions, and conclusions that necessarily follow with the right probability. It’s just critical thinking 101

The proof is in the pudding: For any AI to discover new physics or build technology in reality that actually works... there is BS physics, you have to be extremely truth-seeking

Physics is the law. Everything else is just a recommendation. If your rocket design is wrong, it blows up. Reality doesn’t lie

See 1 related tweets

• @XFreeze: Grok is the least biased, most truth-seeking AI on the planet

built to face reality head-on. Answer...

19. mark_k (Group Score: 52.4 | Individual: 28.4)

Cluster: 2 tweets | Engagement: 582 (Avg: 192) | Type: Tech

Censorship filters are making Google AI increasingly useless. Even worse, this can become dangerous when the model twists, hides, or selectively omits facts instead of simply telling the truth.

This is exactly what @elonmusk has been warning about: AI must be truth-seeking.\n\nQT @romanhelmetguy: It’s very cool that I can’t use AI to locate quotes from ancient historians because Google doesn’t like the content of those quotes. Look at the thinking process. It finds the quote perfectly fine, but then pretends it can’t. AI is censoring Roman history. https://t.co/4595ZOxXq4

See 1 related tweets

• @XFreeze: They're literally teaching AI to LIE

Google's Gemini finds the exact quote from Augustus (via Cassi...

20. alex_prompter (Group Score: 51.9 | Individual: 51.9)

Cluster: 1 tweets | Engagement: 628 (Avg: 71) | Type: Tech

RT @alex_prompter: I almost never share AI generated videos.

As someone who studied film, most viral AI videos make me cringe. They look impressive on the surface but they're hollow. No story structure. No character motivation. No comedic timing. Just pretty visuals with a generic cinematic soundtrack slapped on top and people in the comments losing their minds over rendering quality.

The bar has been on the floor. "AI made this" became the entire value proposition. Nobody asks if it's actually good as a piece of video content.

This one is different.

47 seconds. Three pigeons arguing over a button. The comedic timing actually lands. The voice acting has real intonation, not that flat AI monotone. The camera POV sets up a reveal that pays off. There's a setup, escalation, and a punchline that genuinely caught me off guard.

It's the first AI video I've seen where I forgot it was AI generated and just laughed.

That's the standard. Not "look what AI can do." But "this is actually good and it was made with AI." There's a massive difference between the two and most creators haven't figured that out yet.