A Chart Shows Top AI Models Get About One in Four Facts Wrong Without Search

Summary

A chart shared by Randy Olson examines how often leading AI models give specific facts incorrectly when they answer without search, citing data from Artificial Analysis. The post says that even the best-performing models get about one in four answers wrong. It therefore recommends having a model look up information whenever the accuracy of a fact matters. Olson says the chart was made with evident-charts, described in the post as his open-source agent skill for explanatory charts.