Skip to content
Hacker News front pageAnon84

AI audit of theoretical astrophysics finds 872 errors and traces how they spread through citations

I had AI read my entire field, Theoretical Astrophysics

The project author used AI to map a theoretical astrophysics literature network of 158,059 papers and 554,670 citations, recomputing 659 papers and reporting 872 errors in 433 of them. Later papers carried the errors forward in 367 cases, and 68 criticized them. The deep check covers 4% of the graph's highly cited papers. Two independent AI sessions reviewed the errors; no human has checked them, and a third session agreed with 33 of 40 randomly rechecked errors. The author concedes some results may not hold up.

Why it matters: The author recomputes highly cited papers with AI and follows the citation chain, pairing independent review with error tracking as a way to check the literature.

Read the original ↗Export Markdown