AI System Proves 2008 Quantum Conjecture in 3 Days; Meta Builds Tactile Data Glove

Import AI (Jack Clark) · rss · 2026-11-09

Jack Clark's Import AI 480 covers three stories:

1. AI for Science: a quantum conjecture proved. Physical Superintelligence (PSI) used an internally developed research system running a non-public Anthropic model to prove the Aaronson-Ambainis conjecture, open since 2008, about when quantum computers can be exponentially faster than classical ones. The system produced a complete proof on day three, verified step-by-step by the Lean proof assistant, then went on to improve the result twice and tackle related open problems. Clark frames this as an early taste of the singularity: automated synthetic researchers extending the frontier of knowledge faster than humans can absorb it.

2. Meta's Tactus glove. Meta Reality Labs details a textile-integrated tactile and pose sensing glove for naturalistic manipulation data collection: 1.5mm fingertip stackup, capacitive tactile sensors placed per contact studies, and IMU-fused tracking resilient to hand-object occlusion. Clark's lens: this is exactly the hardware a rogue AI might pay humans to wear for physical-world data and feedback — read papers through "how would an AI use this?"

3. Bootstrapping a lunar industry. Citing a 2016 NASA Kennedy Space Center paper: as little as 12-41 MT of initial mass can seed a self-replicating Moon manufacturing ecosystem. With Starship (100 MT to LEO) plus orbital refueling, landing 12-41 MT on the Moon costs roughly $1.5-4.2B. The plan upgrades through six manufacturing generations — gases/water/crude alloys/solar cells up to locally made electronics — reaching 2000s-level industry within decades.

Original post →

More from Embodied

Embodied channel →