How do we measure if someone is a good mathematician or not?

Letters of recommendation from trusted colleagues have always been essential for evaluating candidates. Hopefully, human recommendations will remain a strong, faithful signal as the utility of other metrics rapidly deteriorate.

This consolidation of power is precisely why there is a need for open model development, and exactly why frontier labs have been lobbying hard to abolish them.