DESI's hint that dark energy is weakening forced an awkward admission: fit it with an independent equation of state and you get phantom behavior — energy density that climbed in defiance of conservation. Vafa's line is the sharp one: computing dark energy independently of dark matter is simply wrong, and the phantom regime is a bookkeeping artifact of that bad assumption. The piece walks the coupled-sector models — Khoury's dark-QCD analogue, Teixeira's energy transfer between the two — that make the anomaly vanish if you let dark matter's mass drift. Worth reading for how a measurement nudged theorists toward a unified dark sector rather than two unrelated mysteries.

A year ago open weights meant a handful of Chinese labs plus Llama. Now the releases come from everywhere, and Brand's taxonomy is the useful part: pure frontier-chasers (DeepSeek, Poolside, Zyphra), sovereign plays (Cohere, Mistral, Trillion), and Big Tech with orthogonal motives — Alibaba upselling closed models, NVIDIA seeding GPU demand. The signal isn't any single model; it's that 'why release open' now has four distinct answers, and the strategic logic behind each tells you who's actually competing on capability versus distribution.

Reanalysis of InSight's seismic record (Nature Astronomy) finds the Martian crust is chemically stratified — low-silica, iron-rich rock below, high-silica above — the signature of molten rock pooling, differentiating, and reprocessing itself. That's the kind of magma recycling we'd assumed needed subduction. If a stagnant-lid world can evolve a complex crust on its own, the habitability filter we apply to small, tectonically quiet rocky bodies may be throwing out planets too early.