Import AI 480: Self-sustaining space economies; AI for science progress, and Meta builds perfect hands for a super…

1. Key Themes

Theme: AI is beginning to autonomously extend the frontier of human knowledge

AI-driven proof of a long-open quantum computing conjecture

Physical Superintelligence used a research system running a non-public Anthropic model to settle a problem open since 2008, with the proof machine-verified.

"The system had a complete proof of the conjecture during the third day of the run, and Lean, a proof assistant in which a computer checks every step, accepted it. We then set the system to work on related open problems. Along the way, the system improved the result twice, and Lean accepted the final version."

"Remarkable to unremarkable" is the new adoption curve

Clark frames machine-generated discovery as following the same normalization path as image recognition.

"In the same way a few years ago it went from remarkable to unremarkable to see computers gain the ability to identify objects in images, we're now in an era where it's rapidly going from remarkable to unremarkable for computers to extend the bounds of human knowledge."

The bottleneck shifts to human comprehension

The scarce resource becomes humans' ability to understand and verify AI output, not the generation of results.

"I believe we're in the early stages of the singularity, defined as the period when the rate of technological and scientific progress exceeds the rate at which we humans can parse out and make sense of it... mostly automated synthetic researchers extending the frontier of knowledge, spitting out results which need to then be understood by humans."

Theme: Distributed training and compute-access constraints are reshaping who can build frontier AI

Cross-continent heterogeneous training works at meaningful scale

Core Automation trained a model across clusters 9,000 miles apart with different chips, parallelism schemes, and batch sizes.

"A mixture-of-experts model with roughly 32B total parameters, 32x sparsity was trained across two clusters about 9,000 miles apart. Notably, these clusters had different accelerators (GB300 and B300), different parallelism schemes and different total batch sizes."

Technique: staggered fragment syncing

Rather than syncing the whole model, the team synced by fragment to reduce communication burden.

"Training was synced every 50 steps: 'Rather than syncing the whole model at once, we synced it fragment by fragment, staggered across each 50-step window.'"

Compute scarcity and infrastructure, not chips, are the binding constraints

The signal is that new entrants can't buy big blocks of compute, and that power and land are the real limits.

"The biggest bottleneck in deployment of accelerated compute is not the chips themselves but power, grid equipment and land... The bottleneck that holds most organizations back is not the raw amount of compute, but the ability to turn islands of compute into one valuable frontier training run."

Orbital compute as a long-horizon ambition

"An ambitious first step would be to pretrain a small model across a few satellites and compare it with a ground baseline. A trillion-parameter orbital run is much further off."

Theme: Off-Earth industrialization is more tractable than commonly assumed

Tiny launch mass bootstraps an industry

A 2016 NASA Kennedy paper argues a self-sustaining lunar industry needs very little initial mass.

"In the baseline case with maximum manufacturing rate, a total of 41 MT is launched to the Moon. In the reduced rate case, only 12 MT is launched to the Moon but the scale of industry is two orders of magnitude smaller so the benefits of the industry are relatively delayed by eight years."

Starship collapses the cost

"When you napkin out the cost of multiple Starship launches and refuelings you can get 12-41MT to the Moon for between $1.5bn and $4.2bn at list prices."

Bootstrap by historical-era technology ladder

"The objective is for the first robotic 'colonists' on the Moon to fabricate a set of, say, 1700's era machines and then to advance them steadily through the equivalent of the 1800's, 1900's, and finally back into the 2000's. We argue that this can be accomplished in just a few decades."

Asteroids are the real prize

"The asteroid belt has everything necessary for it: water, carbon, silicates, metals, oxygen, solar energy... the billion-fold greater resources could allow the industry to expand exponentially until it dwarfs that of the entire Earth within just a few decades."

Theme: Physical-world data collection hardware is becoming strategic infrastructure for AI

Meta's Tactus glove targets the occlusion problem

Cameras struggle during hand-object manipulation, so tactile and pose sensing fills the gap for dexterity data.

"Tactus provides calibrated fingertip force sensing and occlusion-resilient hand tracking in a textile-integrated form factor that preserves natural grasp behavior."

Spec highlights

"(1) a thin, conformal 3D-knit form factor (1.5mm fingertip stackup) designed to minimize interference with natural manipulation, (2) strategically placed capacitive tactile sensors at manipulation-relevant regions informed by empirical contact studies... (3) IMU-fused pose tracking that maintains accuracy through the hand-object occlusions where manipulation data is most valuable."

Humans-as-actuators for AI

Clark notes this hardware is what an AI system might pay humans to wear.

"It's also interesting because this is exactly the kind of hardware an autonomous AI system might want to pay humans to wear so it could gather data and get feedback from the world."

2. Contrarian Perspectives

We are already in the singularity, defined by comprehension lag, not capability

Rather than treating the singularity as a future threshold, Clark defines it operationally and says it has begun.

"I believe we're in the early stages of the singularity, defined as the period when the rate of technological and scientific progress exceeds the rate at which we humans can parse out and make sense of it."

Evaluate research through an "AI-as-adversary/user" lens

A nonstandard analytical habit: ask how an AI would use each paper, analogous to how an intelligence agency would.

"I have begun to find it helpful to view research papers through the lens of 'how would an AI use this', similar to how it is often helpful to view papers through the lens of 'how would an intelligence agency use the insights shared here'."

Space industrialization is cheap relative to its payoff, and the old AI forecasts were too pessimistic

The 2016 paper assumed slow AI progress, yet the plan still worked on a low bar.

"On-going advances in robotic software and artificial intelligence present a very optimistic picture that these levels of robotics will be achieved as Moravec predicted, with lizard-level occurring by 2020, mouse-level by 2030, monkey-level by 2040, and human-level by 2050… Only mouse-level is needed by the end of bootstrapping on the Moon." Clark adds: "By most measures, AI is around monkey-level and approaching human-level now in the 2020s."

3. Companies Identified

Physical Superintelligence (PSI)

• Description: AI science startup
• Why mentioned: Proved the Aaronson-Ambainis quantum computing conjecture using an internal research system running a non-public Anthropic model.
• Quote: "Now, a research system developed internally at Physical Superintelligence (PSI), running a non-public Anthropic model, found a proof for it, settling one of the most fundamental questions in quantum computing."

Anthropic

• Description: AI lab
• Why mentioned: Its non-public model powered PSI's proof system.
• Quote: "running a non-public Anthropic model, found a proof for it"

Meta (Reality Labs Research)

• Description: Big tech R&D division
• Why mentioned: Published Tactus, a textile-integrated tactile and pose sensing glove for manipulation data collection.
• Quote: "Tactus, 'a textile-integrated tactile and pose sensing glove designed from the ground up for naturalistic manipulation data collection'."

Core Automation

• Description: AI startup founded by senior people from OpenAI and Google
• Why mentioned: Published results on heterogeneous, cross-continent distributed training and orbital compute ambitions.
• Quote: "Core Automation - which up until now has been very quiet - is choosing to signal its interest here. This speaks likely to the relevance of distributed training for future large-scale runs, and also to the general lack of availability of big blobs of compute for newer startups to buy."

Starship (SpaceX)

• Description: Heavy-lift launch vehicle
• Why mentioned: Its lift capacity and refueling make lunar industrialization affordable.
• Quote: "Starship has a capacity to lift 100 MT to low-earth orbit, and given multiple refuelings in orbit it's feasible you could then land it on the moon."

NASA Kennedy Space Center

• Description: NASA launch and research center
• Why mentioned: Researchers authored the 2016 lunar bootstrapping paper.
• Quote: "a fascinating paper from researchers with the NASA Kennedy Space Center in 2016 about what it'd take to bootstrap a self-sustaining industry on the moon."

Covenant

• Description: Group behind the largest distributed-trained model referenced
• Why mentioned: Holds the benchmark for largest distributed-trained model.
• Quote: "This isn't the single largest model trained via distributed techniques (that'd be Covenant's 72B model, #449)."

4. People Identified

Scott Aaronson and Andris Ambainis

• Description: Researchers (quantum computing)
• Why mentioned: Authors of the conjecture that was proved.
• Quote: "Open since 2008, the conjecture of Scott Aaronson and Andris Ambainis concerns when quantum computers can be exponentially faster than ordinary ones."

Hans Moravec

• Description: Roboticist/futurist
• Why mentioned: His robotics-intelligence timeline was cited in the NASA report and has since been outpaced.
• Quote: "...these levels of robotics will be achieved as Moravec predicted, with lizard-level occurring by 2020, mouse-level by 2030, monkey-level by 2040, and human-level by 2050."

5. Operating Insights

Use staggered, fragment-wise syncing to make cross-site training practical

For teams without a single giant cluster, the Core Automation recipe shows heterogeneous hardware can be combined.

"Rather than syncing the whole model at once, we synced it fragment by fragment, staggered across each 50-step window."

Start with the minimum viable, sub-replicating system and evolve

The lunar plan generalizes: bootstrap from crude capabilities and climb the ladder rather than designing the full system up front.

"launching a 'sub-replicating system and evolve it toward the self-replication capability'" and "Early generations will 'produce crude "mongrel alloys"'... hardware built out of this stuff will need to be huge to add strength to make up for poor mechanical properties."

Design data-collection hardware around where cameras fail

Tactus targets occlusion, the point where manipulation data is most valuable, a template for robotics data startups.

"This is the kind of thing that cameras can visually report but often leads to sub-par data due to occlusion during manipulation."

6. Overlooked Insights

Human oversight of AI may become a mass profession built around auditable "translucent" models

In the closing fiction, adding more AI layers to analyze AI produced unreadable abstraction, so oversight became a human job.

"if you added other machines they tended towards creating beautiful analysis that was so abstract... that it became impossible for people to follow. So, instead, a human profession had been created around the meta-therapist, in which teams of thousands of people continuously analyzed and critiqued and theorized about the mind of the meta-therapist."

Lunar ice is a finite resource, forcing an early pivot to asteroids

The paper implies a resource clock on the moon-first strategy.

"The ices in the lunar poles are a limited resource so it will be important to move the center of industry to the asteroids as quickly as possible."