Jed Del Rosario

[Jed Del Rosario @Suludnon]

@TMTLongShort Would like you to comment on this.

[SightBringer @_The_Prophet__]

⚡️The frontier is flattening faster than American institutions are emotionally prepared to admit.

A Chinese model reaching the top of a narrow coding benchmark does not decide the AI race.

The significance comes from the shrinking distance between elite systems.

Once capability gaps compress, national advantage moves away from having the single best model and toward producing competent intelligence cheaply, deploying it everywhere, and connecting it to energy, factories, robots, logistics, weapons, finance, and state capacity.

That is where the contest becomes dangerous for the United States.

America still holds extraordinary advantages in advanced chips, cloud infrastructure, research talent, capital markets, software distribution, and frontier laboratories. Those advantages can be squandered through slow permitting, power shortages, fragmented regulation, legal uncertainty, institutional paralysis, and a political culture that treats construction as something to be litigated before it is attempted.

China’s edge comes from a different operating system. It can compress the distance between research, capital, infrastructure, manufacturing, and deployment. A model improvement can move into factories, consumer platforms, surveillance systems, robotics, and state procurement with far less friction. The model may begin slightly weaker and still produce greater national power because deployment compounds faster than benchmark prestige.

The dangerous American illusion is that inventing the best technology guarantees control of the era.

Britain invented enormous portions of the industrial world and still lost relative power. Leadership transfers when one system discovers and another system scales.

The benchmark matters because it shows that raw model capability is becoming globally reproducible. Architecture diffuses. Training techniques diffuse. Synthetic data diffuses. Distillation spreads competence downward. Open models accelerate catch-up. Talent crosses borders. Hardware constraints slow progress, then engineering finds another path.

The moat around intelligence is thinner than the market currently prices.

That means the real moat is the full stack:

abundant energy, semiconductors, data centers, networks, talent, capital, software distribution, industrial integration, robotics, defense adoption, and political permission to build.

A country that regulates the model while failing to build the power plants, transmission, chip fabs, data centers, and robotic factories has misunderstood the battlefield.

The permissionless-innovation argument contains truth, but it remains incomplete. Total deregulation can produce recklessness, concentration, security failures, and public backlash strong enough to trigger heavier control later. The winning system will move fast while imposing narrow, enforceable constraints around catastrophic misuse, espionage, infrastructure security, and military leakage. Broad pre-approval regimes will kill speed. Zero governance will eventually kill legitimacy.

The deeper pattern is that AI is leaving the era of laboratory supremacy and entering the era of civilizational absorption.

The winner will be the society that turns machine intelligence into generalized productive capacity first.

That means millions of businesses using agents.

Factories using autonomous systems.

Governments rebuilding operations around models.

Scientists accelerating discovery.

Power grids expanding around compute.

Children learning beside tutors.

Militaries integrating machine decision systems.

Robots moving intelligence into the physical world.

The country that embeds AI most deeply into ordinary economic life will gain the durable advantage, even if another country wins more benchmark charts.

[Jed Del Rosario @Suludnon]

@georgesiew1984 @kyleichan 'Banning collaboration on a common good like scientific research is considered unethical by most moral standards.'
https://t.co/jvJqWeLDLS

'It is considered an attack by the norms of the modern international community.'

The 'international community' is a polite fiction.

[Jed Del Rosario @Suludnon]

@HighlanderLore I'd get an M60. Where in the rules does it say I have to use a sword?

[Jed Del Rosario @Suludnon]

@georgesiew1984 @kyleichan How is banning collaborations 'attacking' China?

[Jed Del Rosario @Suludnon]

@char15299 @kyleichan @grok The guy blocked me after responding. Lol. But in case you're looking to seethe, consider China's own controls for academic research and ask why it's only the U.S. thtat has to be 'open?' https://t.co/SsBBk8HhkF

[Jed Del Rosario @Suludnon]

@char15299 @kyleichan 'We have highly qualified people who enforce our espionage laws.'

@grok Please list down some of the most prominent industrial espionage carried by the Chinese on U.S. academic institutions.

[Jed Del Rosario @Suludnon]

@rinconhilldad @kyleichan If China can innovate on its own then there should be no problem with decoupling academic institutions. By all accounts it should be the Chinese who want to decouple, not the Americans.

[Jed Del Rosario @Suludnon]

@char15299 @b_bran223 @kyleichan @michaelxpettis India and Japam also restrict/strictly control research collaboration with the Chinese. And yes, you can hide scientific/tech knowledge. It's a modern vanity to think otherwise.

[Jed Del Rosario @Suludnon]

@deepanshu_66194 @Americanist144 @kyleichan You're Indian. Apply your comment to India.

@grok How much control does India exert over its researchers when it comes to collaborating with China?

[Jed Del Rosario @Suludnon]

@georgesiew1984 @kyleichan I think you are being dramatic. China has many issues, but suicide is less likely.

[Jed Del Rosario @Suludnon]

@xh_andrea @michaelxpettis They can rebalance trade or they get trade reduction through tarrifs or decoupling. Those are the options.

[Jed Del Rosario @Suludnon]

@TMTLongShort This is why I think nation-states will become increasingly obsolete in the future. They will exist in name but what is coming next is the return to empire and the buffer states meant to keep them separate.

[Jed Del Rosario @Suludnon]

@Noahpinion If the Japanese, Chinese and Indian states disappeared, there would still be Japanese, Chinese and Indians. If the United States of America disappeared, would there still be Americans?

[Jed Del Rosario @Suludnon]

@andyrobinplays @Jesse8009 @Noahpinion Like Japan?

[Jed Del Rosario @Suludnon]

@kodawolf97 @IrenicCounselor @Inevitablewest Contexts are not universal. What you think is correct is actually very local. What East Asians, South Asians, etc... consider morally acceptable can be radically different.

[Jed Del Rosario @Suludnon]

@TMTLongShort If it's a troll post, where's the China #1 crowd?

[Jed Del Rosario @Suludnon]

@_The_Prophet__ The first part of your post applies to all nations on the world. But I like this framing: Immigration has become America’s demographic bailout.

[Jed Del Rosario @Suludnon]

@_The_Prophet__ To try to throw the spear even further, even if I will most likely fail.

[SightBringer @_The_Prophet__]

⚡️CNBC is now reporting that Anthropic has accused Alibaba of illicitly accessing its AI models.

That is a serious escalation.

This is the AI Cold War moving from chips into model behavior.

Alibaba allegedly accessing Anthropic models means the battlefield is no longer limited to GPU supply. The battlefield is trained intelligence itself: reasoning behavior, coding ability, agent workflows, safety boundaries, tool-use patterns, and the expensive judgment layer embedded inside frontier models.

The strategic mechanism is distillation.

A rival does not need to steal source code to steal capability.

It can query the stronger model at scale, harvest outputs, capture behavior, generate synthetic training data, and use the frontier model as a teacher.

That turns Anthropic’s training cost, research breakthroughs, alignment work, and product behavior into someone else’s upgrade path.

That is why this matters.

A frontier model is condensed compute, data, talent, alignment, reinforcement learning, and operational knowledge.

Unauthorized access becomes a way to launder capability around chip controls.

If China cannot freely access the best chips, one shortcut is to extract behavioral signal from the best models.

That is compute laundering through cognition.

The deeper geopolitical signal is brutal:

Intelligence itself is becoming an export-controlled asset.

The U.S. already treats GPUs as strategic material. Now frontier model access is moving into the same category.

Access to Claude, GPT-class systems, Gemini-class systems, and frontier agent stacks becomes a national-security surface.

This is why model access will keep tightening by geography, identity, endpoint, and organizational affiliation.

The open internet model of “anyone can query the smartest system on Earth” is dying at the frontier.

The future is permissioned intelligence.

China’s incentive is obvious.

Domestic AI progress is real, but dependence on U.S. model behavior and U.S.-controlled hardware is strategically intolerable.

So the channels multiply:

Smuggled chips.
Synthetic data extraction.
Gray-market APIs.
Overseas subsidiaries.
Research fronts.
Talent movement.
Open-source leakage.
Output harvesting.

The Anthropic incentive is obvious too.

It needs to protect the moat, prove security seriousness, and position Claude as national-security-grade infrastructure.

But the underlying logic is real.

This is model espionage entering the mainstream.

The chip war created the hardware border.

The next border is access to frontier cognition.

[Jed Del Rosario @Suludnon]

@shmalkey @zriboua I live in the Philippines. The ideology is premised on the notion that the international system is actually evil and oppressive and would like to flip it around, so to speak. What,s more likely to happen is that it will fragment into blocs, which is what's happening now.