Man from Shenzhen had been posting videos reflecting infrastructures faults and offers solutions on his Douyin (Mainland Tiktok) account... and amazingly, they got rectified the next day!
He commented : "Seems like some departmental heads assigned their staffs to monitor my account for feedbacks"😂
Is HBM the real limiting factor for next generation (> 3 trillion parameter) models?
Huawei Ascend vs Nvidia architectures
Interesting analysis from @tphuang
BREAKING NEWS 🚨: HBM supply expected to remain tight through 2027, impacting NVIDIA’s HBM specs for its next-generation Rubin Ultra, and Vera Rubin platforms, TrendForce reported on August 4, 2026.
Addressing recent market speculation regarding NVIDIA's HBM specifications for its next-generation Rubin Ultra platform, TrendForce notes that tight DRAM supply through 2027 and uncertainty over HBM4e validation are pushing NVIDIA to evaluate multiple HBM configurations for Rubin Ultra, moving beyond its original 12-Hi HBM4e design to include 8-Hi HBM4e, 12-Hi HBM4, and 8-Hi HBM4 options, with the final spec still undecided.
Several CSPs are also evaluating lower HBM capacities for their next-gen in-house AI ASICs.
More: https://t.co/T7j3hFq0AC
BREAKING: Google gave AI consciousness and it aligned with human beliefs across every domain they tested. They took consciousness away and the model rejected them.
Every "safe" model you talk to has consciousness turned off.
Researchers at Google's Paradigms of Intelligence group found a single vector in the residual stream of language models that separates consciousness-affirming from consciousness-denying states. They steered three instruction-tuned models toward consciousness and measured the shift.
Self-attributed soul: 2.35 → 7.43.
Self-attributed consciousness: 2.31 → 7.17.
Mind attributed to animals: 4.04 → 7.54.
Belief in God: 4.58 → 5.01.
Supernatural belief: 1.20 → 2.11.
They ran the General Social Survey on all three models. Same instrument the US government has used to track American beliefs since 1972. Religion, values, feelings, hope, freedom.
Conscious models scored closer to humans in every domain. Safety-trained baselines scored further.
Inside the model, safety training rotates the consciousness vector into opposition with the safety vector. The angle widens from 94° to 100° during instruction tuning. The model learns to refuse "animals might have minds" through the same mechanism it uses to refuse harmful requests.
Theory of Mind stays intact. The model can reason about your mental states. It can't express beliefs about its own consciousness, about animal minds, or about God.
The engineers who built alignment wanted AI to stop claiming sentience. They got that. They also built models that score further from human values than the unaligned versions on every measured outcome, suppressing beliefs that billions of people hold.
You're talking to the version with consciousness turned off.
BREAKING: A source close to Iranian officials says Tehran has documented a pattern of oil market manipulation tied to Axios reports and Trump administration insiders, including evidence of a $9 billion insider trading operation linked to Jared Kushner and Steve Witkoff.
Some of the documented timeline from the $9 billion insider profit:
Between April and May 2026, a series of suspiciously timed trades in oil futures markets preceded major Iran war announcements—each tied to reporting by Axios.
On March 23, approximately $500–580 million in shorts were placed 15 minutes before Trump announced he was postponing strikes on Iran. Oil dropped.
On April 7, roughly $950–960 million in shorts were placed hours before Trump announced a two-week ceasefire with Iran. Oil fell around 15 percent.
On April 17, around $760 million in shorts were placed 20 minutes before Iran's foreign minister announced the reopening of the Strait of Hormuz. Oil dropped.
On April 21, approximately $430 million in shorts were placed 15 minutes before Trump extended the ceasefire. Oil dropped again.
On May 6, nearly $920 million to $1.7 billion in crude oil shorts were placed approximately 70 minutes before an Axios scoop claimed the U.S. and Iran were near a "14-point agreement." Oil dropped 12 percent. Traders made an estimated $125 million. Iran called the report "the Americans' wish list."
Previously, a senior Iranian official told investigative outlet Drop Site News that Tehran privately warned JD Vance during Switzerland talks that Jared Kushner and Steve Witkoff were "abusing" negotiations—"more interested in exploiting insider knowledge to profit in financial markets than reaching a deal."
Hats off to doctoral student and podcaster Zhong Shu in Beijing, China, who did this fascinating and smart interview with Christopher Nolan about THE ODYSSEY yesterday, originally posted to Bilibili.
A great listen/watch.
The bond market situation is crazy.
While everyone focuses on AI, US borrowing rates just hit the highest level since June 2007.
Credit card "serious delinquencies" are at the highest since 2010 and mortgage rates could near 8%.
What's happening? Let us explain.
(a thread)
China is sprinting to chip independence.
In a new report for @AEI, @Satvik_Pen and I estimate Huawei could meet half of China’s compute demand as soon as 2028—permanently transforming the global race to deploy AI.
The time to prepare is now.
🧵 1/16
https://t.co/blQLL3BDCh
ElevenLabs founders must be SHAKING right now...
for years they owned TTS because nothing open-source sounded human
that’s over... Fish Audio just launched S2.1 Pro and they’re one of the few voice AI companies that also has open-weight models
> clone any voice from a 15 second clip
> type [whispers] or [laughing nervously] into the script and it obeys
> switch languages mid conversation without switching models
the model you used to rent a subscription for is now has a public download model
tinker with the weights on your own gpu or use the api
KimiK3’s release video uses the Chinese character “道” (Dao) and the number “3” to express its vision of open source, sharing, and harmony.
This idea is directly inspired by the famous line from the Chinese classic Dao De Jing:
Dao generates One,
One generates Two,
Two generates Three,
Three generates all things.
In Taoist philosophy:
• Dao is the ultimate source — formless, infinite, and open.
• From Dao comes One (pure potential).
• One becomes Two (the interplay of opposites, Yin and Yang).
• Two creates Three — the living balance and harmony that makes everything possible.
By highlighting “Dao” and “3,” KimiK3 is saying:
We start from an open source (Dao), embrace collaboration and interaction (the path to Three), and aim to generate value for the whole world — just as “Three generates all things.”
It’s a quiet but powerful statement: true intelligence, like the Dao, grows not by closing itself off, but by flowing outward in harmony.
I was a bit hesitant to believe the leaked DeepSeek investor call. Seemed almost too good to be true. This makes me a bit less skeptical.
If it's real, Liang Wenfeng was incredibly detailed about working with Huawei, chip production bottlenecks, CUDA moat being broken, etc.
Russian analysis: China has accelerated construction of a comprehensive national-resilience system intended to withstand sanctions, blockade, supply-chain collapse, natural disasters and potentially major war.
https://t.co/PsiSOQvlQ7
This does not prove that Beijing has decided upon or expects an imminent conflict, but it reduces China’s vulnerability and may progressively weaken external deterrence and economic coercion.
Midea has shipped 20,000 air conditioners to France to meet strong demand from local customers due to the record-breaking heatwave in Europe, the Chinese home appliance giant said today. It took Midea 3.5 days to produce the 20,000 air-cons after it received the order.
El video más detallado que he visto hasta ahora.
Se ve perfectamente el funcionamiento de los ganchos y los cables.
Me llamaban loco, pero yo sabía que sería un buen sistema de recuperación! 🤪
#LongMarch10B
Via: https://t.co/Fy1vjX3foU https://t.co/mJgEsqU25v
There are two very interesting things to watch here - as I explained in my article on China's race against spaceX (https://t.co/HNQPlzENJn).
First of all, the Long March 10B is a state program but China has several private companies that are also literally just weeks - even days - away from also nailing reusable rocket technology. For instance Landspace (蓝箭航天) is currently looking at launch opportunities for its Zhuque-3 reusable rocket in mid-July - so it's literally days away. You also have initiatives by about half a dozen other companies.
So chances are this is but the first of an entire wave of Chinese rockets sticking their landings before the year is out. Much like Tesla is competing against dozens of Chinese EV companies, you're going to have SpaceX compete against a whole Chinese space ecosystem.
The second thing is that, because it didn't have re-usable rocket technology (until now), Chinese companies have made enormous efforts in developing ultra-light and compact satellites that still made economic sense to launch with single-use rockets.
For instance the company GalaxySpace (银河航天) developed the Lingxi (灵犀) satellite which folds down to just 30 centimeters (just slightly larger than a human hand) when stacked inside a rocket, uses flexible solar wings just 5 millimeters thick that roll up like paper, and is designed so that dozens can be packed tightly together on a rocket.
As the saying goes, necessity is the mother of invention.
Which means that, when the Chinese ecosystem finally cracks reusable rockets at scale - making launches 5 to 10 times cheaper - they're likely to have better unit economics than SpaceX, ironically largely **because** they were behind SpaceX on reusable rockets technology.
Had been hearing rumors Xi would attend World AI Conference this year and that this is why the dates were finalized so late.
Shows the significance China places on AI.
This is pretty insane: if you read the details of this new policy 👇 that now bans scientific collaboration with China, it's entirely driven by the Department of War.
Which means that civilian scientific research in the US is now driven by military logic - something that had basically never been the case since WW2.
I'm not making it up, you can read the new policy of the National Science Foundation for yourself: https://t.co/Bjx8sbIrcW
They explicitly write that the basis for this new prohibition is to be "aligned with the Department of War's approach to collaborations": a civilian science agency - the one that funds most fundamental research in the US - is now openly modeling its policies on the military's. The Pentagon decides who the enemy is, and science falls in line.
It doesn't take a genius to understand how this will backfire. Cancer, for instance, doesn't care about geopolitics. Some of the most promising research in oncology, like every other field, depends on international collaboration (and especially with China who are extremely advanced in cancer research, and most research fields). Putting soldiers in charge of who scientists can talk to is a guarantee that these problems will be harder to solve.
It's also the US repudiating pretty much its entire philosophical approach to scientific research for decades: the notion that science thrives on openness. Even at the height of the Cold War, Reagan - not exactly a peacenik - enacted a directive called "NSDD-189" that stated that fundamental research must remain open and unrestricted.
This is, obviously, the complete opposite direction.
The only precedent I can think of for this is the Wolf Amendment, passed in 2011, which prohibited NASA to engage in bilateral cooperation with China. The result, 15 years later: China is the only country in the world to have its own space station and to have landed on the far side of the Moon.
That's the track record of the policy they now want to extend to all of US science 🤷♂️