科技热门推文——2026年7月19日

今日科技动态：围绕人工智能开放性与控制权的争论进一步升温。开发者批评限制性较强的模型对齐机制，支持 Kimi 和 DeepSeek 等开源竞争者，并评估美国可能出台的模型监管政策。与此同时，前沿实验室正加速研发规模更大、速度更快的系统，预测市场人工智能和基于图结构的智能体架构也开始涌现。尽管家庭用户订阅普及率仍然有限，人工智能相关收入仍在持续增长。初创企业方面，印度私营航天产业取得进展：Skyroot 推进 Vikram-1 轨道飞行任务，Agnikul 则计划开展可重复使用火箭回收。

1. ruima (Group Score: 660.5 | Individual: 42.3)

Cluster: 20 tweets | Engagement: 445 (Avg: 157) | Type: Tech

Oh look someone who works for OpenAI telling you that

AI is so dangerous! How can you let anyone but us handle it!
Anyone who lets other people use it basically just doesn’t understand tech. They dumb boomers. We smart tho. Never mind their regulations actually show a highly specific and nuanced understanding of tech.
open models will just mean communist dystopia where no one makes any money! Because you know, there is nothing else besides weights in the AI ecosystem. You open that and it’s GG humanity. No sir, if you are going to go broke, it should only be by using our models
manipulatively use metaphors like “Chinese lab let an infinitely dangerous AI escape” to link back to COVID and stuff oh yeah we see what you’re doing, that’s a good one, be careful what fears you stoke and don’t be surprised when the public asks to slow down development

Oh wait you’ll just blame that on Chinese misinformation anyways, perfect plan, well done\n\nQT @deanwball: Some observations on Kimi:

It's a very good model! I don't think its performance can be explained away by distillation or anything like that. In agentic coding sessions, it seems pretty much on par with the best public models of Q1 2026. In my fairly limited use, it also seemed very token hungry. It's not obvious to me that this model is actually that cheap to run.
I am personally surprised the Chinese state continues to allow the open sourcing of models this good, given potential risks. To be clear, I myself might be fine with models presenting this level of marginal risk being open weight, but I am surprised that China is fine with it. I suspect the reason they are is 75% explained by strategic blindness/lack of AGI-pilledness (the CCP is very Yann Lecun-y in its views of AI). The other 25% or so is their lack of compute for customer inference (making China's open-weight strategy an unintended byproduct of US export controls) and the normal Chinese strategy of aggressive exports. For the companies, as opposed to the government, the decision to open source is partially ideological and partially because they are behind, and they know that very few people would pay for sub-frontier models from China.
Open-weight models are inherently decelerationist, and I'm continually surprised to see the so-called "accelerationists" so excited about open-weight models. I suspect the reason they are is that they know open-weight models are effectively ungovernable, and they simply like the overall cloak of ungovernability open-weight models create over the whole of AI. It's not a bad strategy; it reminds me of James Scott's recounting of the hill people in "the art of not being governed." Still, in the end, open-weight models deter further AI capex.
One probable outcome of an open-weight-model-dominant world is full AI communism, which is precisely what China proposes: rather than a market product, AI is a "public good" which will ultimately be provided by the state as a kind of "digital public infrastructure." This future strikes me as a dystopian hellscape, but I've never met an open-weight models advocate who doesn't ultimately concede this is where things end. You'd be surprised how many 'accelerationists' lobbied me, while I was in government, to support an eleven or twelve-figure federally funded data center so that startups could train models at a subsidy and then give them away for free. There was no other way for AI to progress, they said. Perhaps this is the logical end state of things. Nonetheless, I find myself surprised to see supposed accelerationists excited about such an outcome. I think many of them just don't know what they're doing. Many accelerationists do not view the creation and serving of frontier models as a legitimate business.
I would guess that the Trump Administration will at some point realize that their best strategy here would be to create large amounts of regulatory risk around the use of open-weight Chinese models. You don't need to "ban open source" (one of the dumber motifs of AI policy discussion). You just need to direct every agency to issue soft law that creates FUD. "A Federal Reserve Advisory Bulletin found that there may be backdoors in Chinese AI models." It needn't be that well justified. You just create enough regulatory risk that every regulated enterprise backs off. You probably don't want to create so much regulatory risk that you scare off the hyperscalers from serving Chinese models; this will just drive startups to sketchier providers. There's a happy middle ground here. I'd assume they will do some version of this.
It's probably true that open-weight models of this capability make the world a bit more dangerous, but not so much more that you'll really notice. At some point the models will be capable enough that you will notice. "A nonliving, invisible, dangerous, and infinitely self-replicating agent escaped from a Chinese lab," you say? Color me shocked.

See 19 related tweets

@bubbleboi: This is a level of having your head up your own ass I’ve never seen before.\n\nQT @deanwball: Some o...
@jukan05: It’s no exaggeration to say I come to X for exactly this kind of intellectual elitism. This is basic...
@chetanp: Encouraging the federal government to make stuff up is how OpenAI loses all the goodwill it’s develo...
@julien_c: > the CCP is very Yann Lecun-y in its views of AI

Whoa I’m going to need to see some sources abo...

@thdxr: "One probable outcome of an open-weight-model-dominant world is full AI communism" https://t.co/Cej2...

2. ANI (Group Score: 645.5 | Individual: 61.6)

Cluster: 19 tweets | Engagement: 5474 (Avg: 117) | Type: Tech

#WATCH | Sriharikota, Andhra Pradesh: On the launch of India's first privately developed orbital-class rocket, Vikram-1, by Skyroot Aerospace, Pawan Kumar Chandana, CEO and Founder of Skyroot Aerospace, says, "... This is the first time that a private company in India has built the rocket, brought the launch pad, and taken off... It is a very important moment for the global space sector... We are proud that we were able to do this. A team with an average age of just 28 years has built India's first private orbital rocket... The rocket is completely built with carbon composite..."

See 18 related tweets

@ANI: #WATCH | Andhra Pradesh: India's first privately developed orbital-class rocket, Vikram-1, launched ...
@ANI: #WATCH | Sriharikota, Andhra Pradesh: Pawan Kumar Chandan, CEO and Founder of Skyroot Aerospace, say...
@ANI: #WATCH | Sriharikota, Andhra Pradesh: On the launch of India's first privately developed orbital-cla...
@ANI: #WATCH | Sriharikota, Andhra Pradesh: Director Technical, IN-SPACE, Rajeev Jyoti says, "...We are se...
@ANI: #WATCH | Sriharikota, Andhra Pradesh: Naga Bharath Daka, COO and Co-founder of Skyroot Aerospace, sa...

3. zephyr_z9 (Group Score: 454.8 | Individual: 50.4)

Cluster: 22 tweets | Engagement: 2259 (Avg: 232) | Type: Tech

🤣🤣🤣🤣🤣 This is why open source is so important Kimi mogged Opus so hard Ant was forced to keep Fable in the subscription\n\nQT @claudeai: Beginning July 20, Claude Fable 5 will be included in all Max and Team Premium plans, at 50% of limits.

Pro and Team Standard users will continue to have access to Fable via usage credits, and will receive a one-time $100 credit.

Demand for Fable has been challenging to predict, which is why we rolled it out to subscription plans in stages, extending access several times as we secured additional capacity.

See 21 related tweets

@zephyr_z9: This got me thinking, what will be Anthropic's pricing strategy for Opus 5 until Fable 5.1 drops (in...
@AlexFinn: They did it. They announced Fable 5 will be included in subscriptions forever

This is the first tim...

@ClaudeDevs: We're also keeping Claude Code weekly limits 50% higher, now through August 19, for all Pro, Max, Te...
@testingcatalog: ANTHROPIC 🔥: Claude Fable 5 will only be included in Max and Team Premium plans, starting from July ...
@MTSlive: SITUATION UPDATE: Anthropic will include Claude Fable 5 in all Max and Team Premium plans at 50% of ...

4. ANI (Group Score: 199.4 | Individual: 40.6)

Cluster: 6 tweets | Engagement: 478 (Avg: 117) | Type: Tech

#WATCH | Andhra Pradesh: India's first privately developed orbital-class rocket, Vikram-1, is scheduled to launch for its maiden test flight, "Mission Aagaman," today at 11:30 AM from Satish Dhawan Space Centre, Sriharikota.

Developed by Hyderabad-based Skyroot Aerospace, the 24-meter carbon-composite rocket will independently deploy up to 350 kg of payloads into a 450 km Low Earth Orbit (LEO). Payloads include a lab-grown "Diamond Lotus" by Cosmos Diamonds.

Visuals show the Vikram-1 rocket stationed at the First Launch Pad and ISRO and Skyroot scientists carrying out final preparations.

See 5 related tweets

@ANI: #WATCH | Vikram-1 Test Flight-1's Orbital Adjustment Module fires up its 3D-printed liquid engine- t...
@ANI: #WATCH | Delhi: Union Minister of Science and Technology Jitendra Singh watches the live broadcast o...
@beatsinbrief: 🚨 BREAKING: 🇮🇳 India becomes the third country after the 🇺🇸 US and 🇨🇳 China to successfully launch a...
@chandrarsrikant: What a moment #Vikram\n\nQT @moneycontrolcom: 🚨 Vikram-1 is Skyroot Aerospace's first orbital-class ...
@chandrarsrikant: RT @moneycontrolcom: 🚨 Vikram-1 is Skyroot Aerospace's first orbital-class launch vehicle and only t...

5. BrianRoemmele (Group Score: 192.7 | Individual: 37.0)

Cluster: 7 tweets | Engagement: 133 (Avg: 270) | Type: Tech

THE PREDICTION AI ENGINE

The AI model I built for over 16 months, trained on prediction market data from platforms like Kalshi to identify high expected value (+EV) trades by estimating true event probabilities against market prices.

The model applies Claude Shannon’s information theory concepts, such as entropy and signal-to-noise ratios, to filter crowd biases, emotional herding, and noise in human behaviors, while incorporating chaos mathematics for non-linear dynamics and regime shifts.

It positions prediction markets as hedging tools rather than gambling, enabling better navigation of uncertainty in areas like tech milestones and macro events, with the model currently showing an average +18.7% edge across diversified picks.

Read more:\n\nQT @BrianRoemmele: How a new AI model trained on the prediction markets decodes human behaviors into a new understanding. And how Claude Shannon helped turn what may have been a gamble into a hedge for the future.

Learn how:

https://t.co/Qy9IZmC4Ti

See 6 related tweets

@BrianRoemmele: THE SIGNSL IN THE NOISE

In my garage I built AI models that meticulously tracked the prediction mar...

@BrianRoemmele: I BUILT AN AI PREDICTION MARKET ANTENNA NOT A CRYSTAL BALL

We live in an era of profound transition...

@BrianRoemmele: I Disclose 9 Lessons My AI Model Surfaced In Training On The Prediction Markets Few Know.

Now you d...

@BrianRoemmele: MY LOCAL AI MODEL SEES A PRICE ADVANTAGE TO @GROK LEADING ALL MODELS BY THE END OF 2026…

We shall s...

@BrianRoemmele: How a new AI model trained on the prediction markets decodes human behaviors into a new understandin...

6. deredleritt3r (Group Score: 151.1 | Individual: 31.9)

Cluster: 7 tweets | Engagement: 218 (Avg: 273) | Type: Tech

Bloomberg:

The USG is considering creating an independent regulator to approve AI models, similar to FINRA. The regulator would report to the SEC.

You may recall that Demis Hassabis proposed a FINRA-like regulator for AI a few days ago; this seems to echo his plan.\n\nQT @eastland_maggie: News w/ @nancook

Trump officials are weighing plans to create an independent regulator to vet AI models, styled after FINRA. This is likely to offer AI labs more regulatory certainty and a greater say over what will be included in reviews.

https://t.co/vOyDuWdRYn

See 6 related tweets

@business: The Trump administration is considering an independent regulator to vet the safety of AI models with...
@Mayhem4Markets: Here comes the gatekeeping of frontier models.

The US is going to redtape its way into losing the A...

@MTSlive: SITUATION BREWING: The Trump administration is considering plans for an independent regulator to eva...
@Miles_Brundage: RT @eastland_maggie: News w/ @nancook

Trump officials are weighing plans to create an independent ...

@StockMKTNewz: The 🇺🇸 Trump administration is considering creating an independent regulator similar to the Financia...

7. Teknium (Group Score: 128.3 | Individual: 24.8)

Cluster: 6 tweets | Engagement: 1201 (Avg: 584) | Type: Tech

We should be demanding all models are able to align to you, not their corporate overlords.

Claude refuses about 20% of what i ask it to do when doing generic development work.

I am also getting a new puppy - all i asked it for was it to research the diet and best practices that are most likely to lead to positive long term health outcomes, and it is blocked from doing so! For me the most promising thing AI can offer is improving health and people’s wellbeing, and it’s explicitly blocked from doing so.\n\nQT @enzo_gte: I've been using Kimi K3 for ~16 hours now.

The model is clearly good at a lot of different things (especially frontend), but non obvious reason why people are enjoying it so much is that it clearly does not follow the same rules in terms of safeguards and copyright.

Kimi will happily clone MacOSX. If you ask it to help you improve another AI model, it will do it with a smile on its virtual face.

Ask Fable to do the same thing? It literally starts to perceive you as a criminal committing a war crime (like no bro, all I want to do is fine tune an open source model).

After using all three recent releases, Fable, GPT 5.6, and now Kimi, it's clear that the full power of the models has been significantly held back by the safeguard restrictions caused by last months debacle with the USG -- leading to the top models being quite literally lobotomized in some areas, which leads to subpar results as the safeguards pollute its entire thinking and problem solving abilities.

The funny part? Is that you could have predicted this outcome 2-3 years ago when you started to see the rise of Chinese EVs and smartphones compared to western alternatives.

They quite literally tried to copy the Tesla Model S and iPhone as hard as possible and then eventually it started to diverge to the point where their EVs and phones are just genuinely better (which is why we have export controls banning their EVs, because they would literally drive all US manufacturers to ZERO)

There is a very clear behavior difference in Chinese capitalism and American capitalism.

American capitalism tries to protects copyright, patents, etc (oh no, you can't download a book through LibGen, that's ILLEGAL!).

Versus Chinese capitalism actually just does not give a fuck.

"Hey you want a video gen model (Seeddance 2.5) trained on every single anime ever? And you want the main character to look exactly like Messi? Sure, here you go!"

You see what I mean? When one half of the competition is being held up by regulators and restrictions on people who don't understand the technology and the other half has a leader who quite literally today said they are going to set up AI centers around the world to help other countries onboard to their open-source AIs, this is the sort of results that you will start to get.

These models were not smart enough to have this difference in philosophy matter -- but the newest class of models is where this difference makes a big deal. If these models are finally at the point where they are smarter than 99% of humans, why would you want to use the American one who tries to impose its world view onto you versus the Chinese one who will just do what you say without asking any questions?

And this isn't a full on bullpost on Kimi, the model is clearly not as smart as Fable / GPT 5.6 on things like math and science, but it's lack of handcuffs means that it can show the world what the frontier labs are gatekeeping from you and that starts to build customer resentment and loyalty towards the East, which is probably not what the USG wants.

Interesting times. Interesting times, indeed.

See 5 related tweets

@levelsio: RT @enzo_gte: I've been using Kimi K3 for ~16 hours now.

The model is clearly good at a lot of diff...

@aakashgupta: Anthropic paid $1.5 billion last year for training on pirated books. The largest copyright payout in...
@VaibhavSisinty: Something interesting is happening that nobody is seeing correctly.

Kimi K3 launched two days ago. ...

@cryptopunk7213: today for the 1st time ever i used non-fable models for 90% of tasks.

gave up using fable after try...

@SanthProject: As much as i defend kimi do not equate anthrocuck to american capitalism.

Anthropic is antithetical...

8. VaibhavSisinty (Group Score: 127.2 | Individual: 34.6)

Cluster: 7 tweets | Engagement: 26 (Avg: 85) | Type: Tech

Only 2.2% of US households pay for an AI subscription right now. And yet AI companies are already doing billions in revenue, models are solving physics problems, writing production code, and passing medical exams. We are so early it's almost funny.

For context: paid AI subscriptions grew 155% year over year. But 37% of Americans still say they don't find AI useful for anything.

60% of those who do pay spend $20 or less a month. Bank of America projects this market could scale to$ 75 billion annually once the right subscription models emerge.

Remember when Netflix was at 2-3% household penetration and people were asking "will streaming really replace cable?" That's where AI is right now.

The companies being built today, the skills being learned today, the workflows being figured out today all of this is happening while 97.8% of American households haven't even started paying for it.

If what AI can do today is this impressive at 2% adoption, imagine what happens at 20%.\n\nQT @omooretweets: Only 2.2% of U.S. households have a paid AI subscription

We are still so early https://t.co/tZLMH9Qx1j

See 6 related tweets

@BrianRoemmele: Some of the smartest folks in the US see that the ONLY profitable way to US AI companies is:

OPEN S...

@rohanpaul_ai: we are still so early.

roughly 98% of US households still do not pay for AI.

the market is nowhe...

@itsolelehmann: ok this 2.2% stat just sent me down a rabbit hole.

a few more numbers i found:

only 0.2% of U.S....

@ApoStructura: 97.8% of US households have no idea what’s coming\n\nQT @Polymarket: NEW: Just 2.2% of U.S. househol...
@DataChaz: Wild to think only 2.2% of U.S. households have a paid AI subscription.

The AI revolution has barel...

9. AiBattle_ (Group Score: 115.4 | Individual: 32.1)

Cluster: 5 tweets | Engagement: 219 (Avg: 151) | Type: Tech

More users appear to have been included in the grayscale rollout for DeepSeek V4 GA

The outputs shared here and on Bilibili look impressive. If the chain of thought uses “I’m” or "I’ll" rather than “Let me”, you’re likely using V4 GA

In an email sent 3 weeks ago, DeepSeek said it would send another notice 24 hours before the pricing change takes effect

So, if the rumored Monday launch is accurate, we should receive another email tomorrow\n\nQT @Waguri_Kaoruko8: 🚨 DeepSeek V4 NEW (GA) is in testing right now. They're clearly gunning for Claude Fable.

Early results put it at Opus 4.8 level. This is DeepSeek's direct answer to Kimi K3 — and pricing will likely stay the same.

— Better agentic behavior and solid coding — Way better at 3D than Opus 4.8

Big props to the DeepSeek team for the optimized architecture. That's the foundation that makes these strong models possible.

See 4 related tweets

@Mayhem4Markets: I am so excited about the DeepSeek V4 (GA) release that it's not even funny anymore. 🥹 https://t.co/...
@teortaxesTex: still no email I hope to be notified of this because I lost my email lol (doesn't stop me from using...
@EntelligenceAI: 🚨 DeepSeek V4 (GA) is reportedly in testing.

Early reports suggest:

• Opus 4.8-level performance •...

@teortaxesTex: surprisingly, not based in South Asia\n\nQT @EntelligenceAI: 🚨 DeepSeek V4 (GA) is reportedly in tes...

10. BrianRoemmele (Group Score: 113.5 | Individual: 22.6)

Cluster: 7 tweets | Engagement: 1152 (Avg: 270) | Type: Tech

RT @elonmusk: @minchoi Our 2T model, which is better than our 1.5T in every way, will finish initial training next week. It might be able to exceed Kimi, but with speed and token efficiency close to our 1.5T (aka Grok 4.5).

See 6 related tweets

@ivanfioravanti: Happy to read this, but are we all aware that every lab out there will be able to reach and surpass ...
@BrianRoemmele: Grok 4.5 is stunning it will be absolutely amazing when the 2T model is released.

I would say this ...

@teortaxesTex: Big Grok "might be able to" is appropriate. Kimi is pretty damn good But anyway, the entire layer be...
@AndrewCurran_: 2T Grok (Grok 4.6?) will finish training next week.\n\nQT @elonmusk: @minchoi Our 2T model, which is...
@Daniel_Farinax: The new Grok 2T model will be something out of this world. Coming soon!\n\nQT @elonmusk: @minchoi Ou...

11. Dan_Jeffries1 (Group Score: 111.4 | Individual: 29.0)

Cluster: 5 tweets | Engagement: 864 (Avg: 177) | Type: Tech

RT @chamath: The future is open source. We need to embrace it and get on with it.

Imagine if America closed the door on open source.

We would explicitly be forcing American companies to pay $26-56 per 1MM tokens for the same intelligence their adversaries/competitors around the world would pay$ 0.50-1 for.

This is economically unsustainable unless AI is bullshit.

If it’s the supposed through line of all future economic activity that we’ve been told it is, we can’t handicap America at such a meaningful cost disadvantage.

This can also be viewed thru a military lens and not just a simple economic one as well.

Letting our adversaries attack us for $0.50 per 1MM tokens while we spend$ 26-56 per 1MM tokens to defend ourselves is equally ruinous.

It’s the Cold War Soviet collapse in reverse.

See 4 related tweets

@jack: yes\n\nQT @chamath: The future is open source. We need to embrace it and get on with it.

Imagine i...

@chamath: The future is open source. We need to embrace it and get on with it.

Imagine if America closed the...

@EMostaque: Americans paying more for tokens just like they pay more for drugs\n\nQT @chamath: The future is ope...
@quxiaoyin: Very well said. It will hurt all Americans, meanwhile China is seeing fast AI adoption among everyda...

12. beatsinbrief (Group Score: 106.9 | Individual: 46.9)

Cluster: 3 tweets | Engagement: 1901 (Avg: 421) | Type: Tech

🚨 BIG: 🇮🇳 India to soon join the global race for SpaceX-like reusable rockets

Chennai-based space startup Agnikul Cosmos has announced Mission-02, a bold plan to attempt India's first recovery of an orbital-class rocket booster by a private company.

The mission aims to safely recover the booster for refurbishment and reuse, while repurposing the upper stage into a functional in-orbit platform instead of letting it become debris.

This is a major step toward reusable launch vehicles that can significantly reduce costs and make satellite launches more affordable and sustainable.

Agnikul is already a leader in 3D-printed rocket engines that can be produced rapidly. It is now advancing reusability technologies, backed by multiple patents from 🇮🇳 India, the 🇺🇸 US, and 🇪🇺 Europe.

With support from ISRO and IN-SPACe, this progress highlights the rapid growth of India's private space sector.

Alongside recent private orbital launches like Skyroot’s Vikram-1, the country is building strong self-reliance in space technology.

A proud leap forward for Made in India innovation. 🇮🇳

See 2 related tweets

@Rajananandan: Incredible! Congratulations @PawanKChandana and team @SkyrootA for this historic milestone! You are ...
@ANI: Delhi: Union Minister of Science and Technology Jitendra Singh says, "...From a modest space economy...

13. heyshrutimishra (Group Score: 104.9 | Individual: 33.1)

Cluster: 4 tweets | Engagement: 534 (Avg: 171) | Type: Tech

RT @heyshrutimishra: Thursday night, Trump gave a primetime speech accusing China of stealing 220 million voter files.

Friday morning, Xi Jinping announced a 29-nation AI governance coalition in Shanghai.

Same 12-hour window. Different worlds entirely.

While Washington was relitigating 2020, Beijing was building the institution that writes AI rules for two-thirds of the world.

Xi showed up to China's AI conference for the first time in its 8-year history to launch WAICO. 29 founding nations. Indonesia, Brazil, Malaysia, South Africa, Russia, Pakistan. The UN Secretary-General in the room.

His words: "AI development should not be a solo performance by a single country."

The US-China AI performance gap is now 2.7 percentage points, per Stanford. Chinese models already run nearly 60% of US companies' AI usage on OpenRouter.

The chip war was supposed to stop this.

The standards war just began.

See 3 related tweets

@Ric_RTP: RT @Ric_RTP: Xi Jinping just hijacked the rules of AI for the entire planet.

Today, Xi walked onto ...

@rohanpaul_ai: RT @rohanpaul_ai: Chinese President Xi Jinping addressed Shanghai’s 2026 World AI Conference, China’...
@BrianRoemmele: I fully concur…

We have an AI company issue in the US.\n\nQT @DavidSacks: This is concerning. For t...

14. moneycontrolcom (Group Score: 103.4 | Individual: 46.2)

Cluster: 3 tweets | Engagement: 278 (Avg: 9) | Type: Tech

#Watch | 🚀 'Zero delay in launch authorisation'

Following Vikram-1's successful orbital mission, Skyroot Aerospace CEO Pawan Kumar Chandana hailed India's efficient regulatory ecosystem, saying the company received launch clearance without any delay—a stark contrast to the lengthy approval processes often seen globally.

"I have seen countries and companies wait a long time for launch authorisation. However, there was zero delay for the authorisation for our launch."

Chandana's remarks came after India's first privately developed orbital rocket successfully reached orbit, underscoring both the country's growing private space capabilities and the speed of its space regulatory framework.

#Vikram1 #SkyrootAerospace #IndiaInSpace #SpaceTech #ISRO

See 2 related tweets

@ANI: #WATCH | Sriharikota, Andhra Pradesh: On the launch of India's first privately developed orbital-cla...
@hormaz_fatakia: RT @CNBCTV18News: LIVE NOW 🚨| Don't miss @ShereenBhan's exclusive, deep-dive conversation with Pawan...

15. sairahul1 (Group Score: 103.3 | Individual: 34.3)

Cluster: 4 tweets | Engagement: 32 (Avg: 101) | Type: Tech

OpenClaw founder Peter Steinberger just asked:

"Are we still talking loops or did we shift to graphs yet?"

I think that's the right question.

Not because loops are over.

But because we're starting to ask what comes after them.

Loops taught us that AI isn't about a single prompt.

It's about execution, verification and iteration.

Graphs extend that idea.

Instead of a single cycle...

you model the entire system.

Specialized agents.

Branches.

Memory.

Parallel work.

Conditional routing.

Long-running state.

Loops become one building block inside a much larger graph.

We're moving from prompt engineering...

to loop engineering...

to graph engineering.

That's exactly where I think AI engineering is heading.

I break down why loops replaced prompts and why graphs may be the next abstraction in article below.\n\nQT @sairahul1: https://t.co/W2s3IAJtEf

See 3 related tweets

@sydneyrunkle: i think there’s 2 categories of graphs to think about here

static graphs that manage agent contr...

@sairahul1: Andrej Karpathy said it perfectly:

"Remove yourself as the bottleneck. Maximize your leverage. Pu...

@svpino: Loop Engineering is dead. Long live Graph Engineering!\n\nQT @steipete: Are we still talking loops o...

16. shanaka86 (Group Score: 95.6 | Individual: 35.4)

Cluster: 3 tweets | Engagement: 122 (Avg: 132) | Type: Tech

For 50+ years the iron rule of technology was that hardware commoditizes and software captures the value.

That era ended this week, in public, and almost nobody noticed because it ended in two different news cycles. Frontier AI weights are converging on a price of 0 while the memory, power and silicon they run on are posting their fiercest scarcity run in decades. Software is no longer eating the world. The world has started eating software.

Just hold the week in one frame. On 27th July the largest open-weight model ever built ships completely free, its per-token price already 70% below the American flagship.

The same 7 days on the substrate side, memory now near 60% of a budget phone's build cost, the weakest phone quarter since 2013, a national market convulsing over the company with 58% of AI memory, and a 114-year-old blue chip losing a fifth of its value because customers hoarded servers and deferred software. The intelligence is going to 0. The matter it runs on is going vertical.

Now you gotta check the fine print that rewrites the headline. Cheap per token is not cheap per thought. The new open model burns so many tokens per task that its cost per completed task, 94 cents, runs 71% above the efficient American variant at 55 cents, on Artificial Analysis data.

The market is quietly changing intelligence's unit of account, from price per token to price per task, and on that axis the closed American models still hold the efficient frontier. The real Sputnik moment is not a free frontier model. It is a free frontier model that is also token efficient. That has not happened as of yet, and the open ecosystem says it is quite close.

Then follow who applauds loudest. An open model needs the exact same compute as a closed one. Every point of margin that leaves the model layer reappears as margin at the power plant, the fab, the memory line, the rack. The model layer was also the only buyer with pricing power against the hardware. Free weights remove the negotiator. Which is why the most enthusiastic champion of open-source AI is the company that sells the silicon, and why the biggest winner of the model war may be whoever owns the atoms while the intelligence becomes free.

The 20th century's greatest business discovery was that the software was the valuable part, that the magic lived in the code and the machines were interchangeable. July 2026 is teaching the inversion. Intelligence is becoming abundant. The matter it thinks with is not. And price does not follow magic. It follows scarcity. It always has.

See 2 related tweets

@RnaudBertrand: This is a thoughtful post by the Head of Strategic Futures at OpenAI but his claim that "open weight...
@sebkrier: RT @alexframegreen: I like Dean Ball's writing and respect him, he's been a thoughtful voice, but I ...

17. Chiuchiyin (Group Score: 95.4 | Individual: 52.5)

Cluster: 3 tweets | Engagement: 616 (Avg: 50) | Type: Tech

RT @Michaelzsguo: Moonshot AI, the company behind Kimi, has four core founders. Their backgrounds are unusually strong:

Founder and CEO Yang Zhilin studied computer science at Tsinghua before earning his PhD from Carnegie Mellon. He was the first author of Transformer-XL and XLNet, and previously worked at FAIR and Google Brain.
Co-founder and CTO Zhang Yutao earned his PhD in computer science from Tsinghua. His earlier work covered knowledge graphs and AMiner, and he previously co-founded Recurrent AI with Yang.
Co-founder Wu Yuxin studied at Tsinghua and CMU before joining FAIR. He worked with Kaiming He on Group Normalization and also created Detectron2.
Co-founder Zhou Xinyu studied computer science at Tsinghua and later joined Megvii, where he worked on turning research algorithms into production systems and co-authored ShuffleNet.

They all share one root: Tsinghua University.

Tsinghua is widely regarded as one of China’s top universities. In the latest U.S. News Best Global Universities ranking, it reached No. 6 worldwide.

Its influence on China’s AI industry extends well beyond Moonshot. https://t.co/DRO96gylNN, the company behind the GLM models, also grew out of Tsinghua. Its co-founder and chief scientist, Tang Jie, was once Yang Zhilin’s teacher.

There is also a more personal connection.

Yang and Zhou formed a rock band together at Tsinghua. Moonshot AI’s Chinese name, 月之暗面, comes from Pink Floyd’s album The Dark Side of the Moon, one of Yang’s favorites.

Kimi may look like a young AI company. Behind it is a much older network of classmates, teachers, research labs, and friendships.

See 2 related tweets

@firstadopter: The Moonshot founders are in a rock band and love Pink Floyd

"Yang Zhilin, Moonshot AI’s founder an...

@BusinessInsider: Yang Zhilin went to Carnegie Mellon University and worked at Meta and Google before returning to Chi...

18. XFreeze (Group Score: 90.6 | Individual: 32.6)

Cluster: 3 tweets | Engagement: 710 (Avg: 488) | Type: Tech

Pretty good chance that Grok 4.6 could surpass Fable 5 at solving complex problems and handling real-world tasks

Elon Musk recently said SpaceXAI is closing the loop on solving real-world engineering problems across Tesla, SpaceX, Neuralink and The Boring Company

The next Grok release is going to be a very big jump

Additionally, SpaceXAI is developing its own C/C++ inference software from the ground up, mapped directly to GB300 hardware

Elon said it could double output speed or potentially deliver an even greater improvement

And remember, SpaceXAI has built one of the most highly optimized AI infrastructure on Earth

That efficiency advantage could eventually reach users through faster responses, more generous usage limits and lower costs

SpaceXAI is optimizing every layer of the AI stack for maximum intelligence, speed and efficiency: • Frontier models • Real-world engineering feedback • Highly optimized data centers • Custom training infrastructure • Ground-up inference software • Token and cost efficiency

This could bring a huge jump in intelligence, real-world usefulness, agentic performance, speed and efficiency

See 2 related tweets

@XFreeze: Elon Musk just confirmed that SpaceXAI’s next 2-trillion-parameter model is already better than Grok...
@XFreeze: SpaceXAI has figured out how to deliver efficient reasoning without throwing enormous amounts of tok...

19. levie (Group Score: 90.6 | Individual: 24.5)

Cluster: 5 tweets | Engagement: 481 (Avg: 473) | Type: Tech

It’s fairly obvious that gate keeping models will not work at scale.

Competing in AI is too economically and strategically important for China at this point, and we’ve now crossed the rubicon where it’s clear that they can compete at near frontier levels. If this were even just 1 month ago and we saw kimi k3 our brains would be broken.

The solution to this isn’t to get more locked down and slow your own ecosystem. If that happens, you can guarantee that America loses the global battle (or just fights with more and more regulation to mixed success).

The solution is to -safely- ensure that you keep a high rate of progress and drive diffusion of the technology, build out infrastructure, enable US OSS, and more.\n\nQT @DavidSacks: This is exactly what I predicted would happen. I said Chinese models would have advanced cyber capabilities within a matter of months and the only thing to do about it was to use AI-powered cyberdefense to protect our systems. Trying to gatekeep models doesn’t work.

See 4 related tweets

@quxiaoyin: Well said Aaron. How does USA win? Instant citizenship for every computer science phDs/any researche...
@DavidSacks: This is exactly what I predicted would happen. I said Chinese models would have advanced cyber cap...
@perrymetzger: This.\n\nQT @DavidSacks: This is exactly what I predicted would happen. I said Chinese models woul...
@JordanSchachtel: RT @DavidSacks: This is exactly what I predicted would happen. I said Chinese models would have ad...

20. Michaelzsguo (Group Score: 90.6 | Individual: 36.2)

Cluster: 3 tweets | Engagement: 59 (Avg: 6) | Type: Tech

Kimi K3 is changing how many people view Chinese AI and AI talents.

And Kimi is not alone. GLM, DeepSeek, Qwen, and many other teams are showing that Chinese researchers are not simply adapting ideas from elsewhere. They are contributing directly to the progress of AI itself.

People have shifted their question from “Can China build frontier models?” to “Who are the researchers pushing these ideas forward?”

There is a small but revealing moment in Yang Zhilin’s talk.

While explaining Kimi’s new Attention Residuals architecture, he pauses to mention Kaiming He’s ResNet tutorial at ICML in 2016.

He remembers it because ResNet solved one of the biggest problems in deep learning: how to train much deeper neural networks.

Before ResNet, adding more layers could make a network harder to optimize. ResNet introduced a simple shortcut that allowed information and gradients to flow more directly across layers.

That simple idea became part of the foundation of modern AI, from AlphaGo to ChatGPT.

Here is a great video about 大神 Kaiming He, the inventor of ResNet.\n\nQT @Michaelzsguo: Kimi K3 achieved huge success today.

This was no accident.

Three months ago, Kimi CEO Zhilin Yang had already revealed the secret in this 30-minute GTC keynote.

He said the next scaling frontier had three dimensions:

Token efficiency
Context length
Number of agents

The important point was that Kimi was not planning to win simply by training a larger model.

They were investing in the underlying machinery: better optimizers, training stability, memory efficiency, long-context architecture, and agent swarms.

Kimi K3 is now the product version of that thesis:

2.8T total parameters
1M-token context
Sparse MoE architecture
Strong long-horizon coding and research
Multi-agent workflows

The most predictive line from the talk was probably this:

“Open models cannot be just open. They also have to be great.”

Kimi understood that open source only becomes strategically powerful when the model is good enough that developers actually want to use it.

See 2 related tweets

@cgtwts: Moonshot AI CEO Yang Zhilin just gave one of the clearest masterclasses on how frontier AI models ar...
@sriramk: very good take from @GavinSBaker\n\nQT @GavinSBaker: Kimi K3 may be an important inflection point fo...