Field notes · In the news
A model trained on four million fossils says three of the big five were slow
Nature news, 20 Oct 2026. nature.com/articles/d41586-026-03316-4
The big five mass extinctions anchor every paleobiology syllabus, and the standard picture is a cliff in diversity at each boundary. Nature reports a machine-learning model trained on close to four million fossil occurrences that reconstructs Phanerozoic diversity at finer resolution than range-through counts allow, and its verdict is that three of the five were drawn out, with diversity sliding for millions of years before the boundary rather than collapsing at it. Expect the argument to centre on the Signor-Lipps effect: incomplete sampling makes abrupt events look gradual, and a model trained on the record inherits its gaps unless it corrects for them.