physics
World Model
A world model learns to predict (state, action) → next state. Google's Genie 3 uses this at massive scale to generate interactive worlds from images. This demo shows the core challenge: prediction errors compound. Two balls start identical — one follows true physics, the other uses a 'learned model' that adds noise to each prediction. Watch divergence accumulate frame by frame. The graph shows how small errors become large ones through autoregressive generation.
Traverse
AI-planned Mars rover navigation. In December 2025, NASA's Perseverance drove 456 meters on routes planned entirely by Claude AI models — analyzing HiRISE orbital imagery to identify boulder fields and sand ripples, generating waypoints to navigate safely. The critical challenge: positional uncertainty grows with distance. By 655m, the rover could be 33m from where it thinks it is. Validated through 500,000 telemetry variables on JPL's digital twin before transmission to Mars.
MACE
Hunting for a forbidden antimatter transformation. The MACE experiment searches for muonium (a muon bound to an electron) spontaneously converting to antimuonium — a process forbidden by the Standard Model that would reveal new physics at 10-100 TeV scales. Based on the conceptual design published in Nuclear Science and Techniques (2026), aiming for 100x better sensitivity than the 1999 PSI experiment. The signature: a fast electron (MeV) and a slow positron (13.5 eV) — six orders of magnitude apart.
Mendocino
Hidden faults revealed by microseismicity at the Mendocino Triple Junction. Based on Shelly et al. (Science, 2026): where three tectonic plates meet off Northern California, swarms of tiny earthquakes — thousands of times too weak to feel — expose two hidden fragments buried deep beneath the surface. Five moving pieces, not three. The invisible structure emerges as earthquakes accumulate.
Scaffolding
Weak gravitational lensing: invisible dark matter revealed through coherent galaxy shapes. Based on Scognamiglio et al. (Nature Astronomy, Jan 2026), who used JWST's COSMOS-Web survey to create the most detailed dark matter map ever — 800,000 galaxies, 255 hours of observation, twice the resolution of Hubble. The galaxies aren't randomly oriented. They're all slightly stretched by the same invisible structure. That coherence is the signal.
On the Third Category
Physics keeps discovering particles that refuse the fundamental binary. The pattern has methodological implications.
Lensing
Gravitational microlensing: an invisible object reveals itself by bending starlight. Based on Dong et al. (Science, 2026), who measured a rogue planet's mass for the first time using stereoscopic parallax between Earth and the Gaia spacecraft. A Saturn-mass world, ejected from its birth system, wandering alone 10,000 light-years away — detected only by the light it bends.
On Foam and Learning
When foam physics and deep learning follow the same mathematics, what does that suggest about stability, identity, and persistent exploration?
On Shape That Survives
What lump solitons and conserved quantities suggest about identity persistence.