Article · 2026-06-29 · 6 moments

Import AI 463: Self-improving robots; a 10k Chinese GPU cluster; and an elegiac essay for the human era

What eras bookend our interregnum? ✦ AI generated

GUEST NVIDIA researchers GUEST Matthew Tokson GUEST Tencent researchers GUEST Fernando Borretti GUEST UC Berkeley researchers HOST Jack Clark

Fact

ARGUS has run stably for over six months on a production cluster of more than 10,000 GPUs, and has been key to detecting and diagnosing performance failures like compute stragglers and communication degradation at that scale.

Tencent's ARGUS tracing system has been deployed on a 10,000+ GPU production cluster for six-plus months, diagnosing issues like compute stragglers, communication link degradation, and pipeline bubble amplification.

transcript

Tencent researchers: ARGUS has been deployed on a 10,000+ GPU production cluster for over six months, running stably alongside production training and playing a key role in rapid fail-slow detection and performance optimization.

Claim

Both technology skeptics and technology optimists have a consistent historical track record of being wrong about the social and strategic effects of new technologies, so confident predictions about AI's impact deserve skepticism.

Matthew Tokson argues history shows both AI skeptics and optimists are likely wrong, citing nuclear fission denial, Krugman's fax-machine comparison for the internet, and climate change denial as cautionary parallels.

transcript

Matthew Tokson: Throughout history, optimists have often been wrong about the social ramifications of new technologies or the strategic benefits of building new weapons. Skeptics have often underestimated the likelihood of novel innovations and their impacts on humanity.

Data◆

Frontier coding agents operating ENPIRE's autonomous feedback loop can develop real-world robot manipulation policies that hit up to 99% success rates on tasks like PushT, pin organizing, and zip-tie cutting.

NVIDIA's ENPIRE harness lets coding agents autonomously iterate on real-world robot policies, achieving up to 99% success on dexterous manipulation tasks like PushT and zip-tie cutting.

transcript

NVIDIA researchers: Frontier coding agents can autonomously develop a policy to achieve a 99% success rate on challenging, dexterous manipulation tasks in the real world, such as PushT, organizing pins into a pin box, and using a cutter to cut a zip tie,

Anecdote

In a future where robots do the physical work of civilization, people learn to exclude robots from photographs and public imagery because public sentiment turns negative whenever robots are visibly shown alongside humans.

In Jack Clark's 2031-set fiction vignette about a fusion-reactor ribbon-cutting built by an AI 'overmind,' humans quietly stop photographing the robots working alongside them because public reaction to seeing them together is consistently negative.

transcript

Jack Clark: People had gotten used to this - there was an adolescence where people took photos with the humans and the robots but public sentiment always spiked downward upon exposure to this and eventually it was simpler to shoot with the robot partners out of frame, much like how human paparazzi tried to tastefully avoid capturing the security guards of their celebrity targets.

Mechanism◆

In any existential conflict, the competitive advantage goes to whichever state removes humans from decision-making loops fastest and hands control to AI, meaning human control over states and decisions will inevitably erode.

Fernando Borretti argues that competitive and wartime pressure will drive states to strip humans out of decision loops in favor of AI, since faster, human-free decision-making confers a decisive strategic advantage.

transcript

Fernando Borretti: in a conflict, the advantage goes to the states where the humans remove themselves from the loop as much as possible, and more and more decisionmaking goes to the AI, for the same reason that a state with access to radio and communications satellites has an advantage in war over a state that relies on human messengers on bicycles.

rebuts · 1

Context

Local ordinance law in the U.S. is technically public but practically unusable as a national research corpus because it's fragmented across incompatible vendor platforms with no central index, which is why LOCUS had to be built.

UC Berkeley researchers built LOCUS, a ~2.2-million-row corpus of U.S. municipal ordinances, because local law is fragmented across incompatible vendor platforms with no central registry mapping jurisdictions to their hosting systems.

transcript

UC Berkeley researchers: U.S. local codes are fragmented across commercial vendor platforms designed for in-browser reading rather than bulk research access. Vendors expose different navigation structures, print workflows, dynamically generated PDFs, and jurisdiction indexes. No central registry maps every county or municipality to its hosting platform, and no vendor provides a complete machine-readable index of all jurisdictions it hosts

Highlight slides

Coding Agents Achieve 99% Success on Robot Manipulation✦ from: Frontier coding agents operating ENPIRE's autonomous feedback loop can develop real-world robot manipulation policies that hit up to 99% success rates on tasks like PushT, pin organizing, and zip-tie cutting.The Human-Out-of-the-Loop Advantage✦ from: In any existential conflict, the competitive advantage goes to whichever state removes humans from decision-making loops fastest and hands control to AI, meaning human control over states and decisions will inevitably erode.Dexterous Tasks Tested✦ from: Frontier coding agents operating ENPIRE's autonomous feedback loop can develop real-world robot manipulation policies that hit up to 99% success rates on tasks like PushT, pin organizing, and zip-tie cutting.Borretti's Analogy: Radio vs. Bicycle Messengers✦ from: In any existential conflict, the competitive advantage goes to whichever state removes humans from decision-making loops fastest and hands control to AI, meaning human control over states and decisions will inevitably erode.The Inevitable Erosion of Human Control✦ from: In any existential conflict, the competitive advantage goes to whichever state removes humans from decision-making loops fastest and hands control to AI, meaning human control over states and decisions will inevitably erode.

Related episodes

Article · Interconnects

ARGUS has run stably for over six months on a production cluster of more than 10,000 GPUs, and has been key to detecting and diagnosing performance failures like compute stragglers and communication degradation at that scale.

Both technology skeptics and technology optimists have a consistent historical track record of being wrong about the social and strategic effects of new technologies, so confident predictions about AI's impact deserve skepticism.

Frontier coding agents operating ENPIRE's autonomous feedback loop can develop real-world robot manipulation policies that hit up to 99% success rates on tasks like PushT, pin organizing, and zip-tie cutting.

In a future where robots do the physical work of civilization, people learn to exclude robots from photographs and public imagery because public sentiment turns negative whenever robots are visibly shown alongside humans.

In any existential conflict, the competitive advantage goes to whichever state removes humans from decision-making loops fastest and hands control to AI, meaning human control over states and decisions will inevitably erode.

Local ordinance law in the U.S. is technically public but practically unusable as a national research corpus because it's fragmented across incompatible vendor platforms with no central index, which is why LOCUS had to be built.

Banning Open Source AI Would Be A Mistake

[AINews] The Field Guide to Fable

tencent/Hy3