groundstate

Friday, August 14, 2026

The Index

◈ Science

Magnus Mountains: Neutron Stars as Continuous Gravitational Wave Sources

arXiv · Aug 14 · ~15 min

A fresh preprint works out the mechanics of what Gangwar and Jones call a “Magnus mountain”: a deformation of a spinning neutron star’s elastic crust driven by its own superfluid vortices. As the star spins, the interior superfluid organizes into quantized vortices that pin to lattice defects in the crust. Any asymmetry in that pinning distribution around the spin axis lets the Magnus force—the same lift mechanism that curves a curveball—push the crust outward non-uniformly. The result is an asymmetric lump: a “mountain.” They solve the coupled elastic-fluid equations self-consistently and find mass quadrupoles as large as ~10⁻⁵ in dimensionless units, limited by crust breaking strain and pinning strength. That’s large enough to be a promising target for continuous gravitational wave searches with current and next-generation detectors—not a merger chirp but a persistent sinusoidal signal, slowly drifting as the star spins down. The physical picture is genuinely clean: quantum condensate mechanics in nuclear-density matter directly producing a macroscopic GW signature. The paper is dense, but the mechanism is worth understanding.

⬡ AI & Product

Gemini 3.7 Flash: GDM’s Fastest Iteration Yet

Google · Aug 13 · ~5 min

Three weeks after 3.6 Flash—which was itself a fast release—Google DeepMind ships 3.7 Flash with a jump that’s hard to ignore: FrontierCode 1.1 Main goes from 34.4% to 43.6%, DeepSWE v1.1 from 49.0% to 65.3%, WebDev Arena Elo from 1538 to 1588, and the price drops to half of 3.6 Flash per token. The product story is agents: better multi-step tool use, more disciplined planning, improved instruction-following when intent is ambiguous. Google Spark—the 24/7 personal AI agent that debuted at I/O—gets upgraded to 3.7 Flash immediately. The strategic context matters: 3.5 and 3.6 Flash had visibly slipped behind Claude 4.8+ and GPT 5.5+ on competitive leaderboards. This release closes much of that gap and does so at a three-week cadence that signals a different pace of iteration than the quarterly big-drop dynamics of a year ago. GDM isn’t conceding the Flash tier.

● Wildcard

The Forgotten Psychedelic Gets Its First Brain Scan

Nautilus · Aug 13 · ~7 min

Mescaline is the structural outlier among classical psychedelics: a phenethylamine (chemically closer to dopamine than serotonin) where psilocybin and LSD are tryptamines. Despite producing subjectively similar effects at matched dose, it engages a different pharmacological profile—and it’s been almost entirely skipped by the neuroimaging wave, a gap partly explained by DEA scheduling and the ceremonial context of peyote complicating IRB approvals. New imaging work is among the first to actually scan it. The finding that it travels a distinct neural path from psilocybin—different circuits at matched subjective intensity—forces a useful question: if two structurally different compounds with different receptor profiles converge on similar phenomenology, what are the “psychedelic state” biomarker theories actually tracking? The pharmacological diversity here is an asset, not a complication. Worth reading for anyone thinking about what the psychedelic neuroscience boom has and hasn’t actually shown.

What would a black-to-white hole explosion actually look like?

Loop quantum gravity predicts that black holes can tunnel into white holes, bypassing the classical singularity through quantum effects. It’s been a theoretical fixture for years—the open question is observability. A new preprint (arXiv:2608.12380) works out the electromagnetic signature in earnest: fast ejecta sweeping the ISM produce synchrotron + inverse-Compton emission peaking in MeV–GeV with no achromatic jet break, evolving on sub-second to tens-of-seconds timescales. That’s distinguishable on paper from short GRBs and magnetar flares. The approximations are significant and the authors say so. But treating LQG predictions as actual observational hypotheses—rather than leaving them forever in the theory tier—is the right move, and LSST and future MeV survey telescopes might realistically test it. → arXiv:2608.12380

A pelican, three browsers, two different results

Simon Willison tested Gemini 3.7 Flash by asking it to generate an SVG of a pelican riding a bicycle at high, medium, and low thinking effort. The high-effort version is genuinely good. The wrinkle: Safari renders it correctly—pelican, bicycle, the works. Chrome and Firefox silently omit the pelican, because they handle empty SVG <filter> elements more strictly than Safari does. A technically valid output, interpreted differently by three different runtimes—a small illustration of a recurring problem: “correct” can be context-dependent in ways that neither models nor humans producing outputs know to anticipate. → simonwillison.net