AI audited 158,059 astrophysics papers and found 872 errors—most still being copied

Casey Handmer, a Caltech PhD, used AI to screen 50,269 full-text papers and recompute 659 of the most-cited ones in theoretical astrophysics. It verified 872 errors in 433 papers; 367 papers copied an error, while only 68 criticized one. Half of copied errors are still being copied seven years later, and peer review removed only about one in five. The whole audit cost roughly $3 per verified error.

A field outgrows its errors. It rarely corrects them.