Random note: I’ve been squinting at the statistical validity of published AI benchmarks in the last few days, and while I vaguely believe the mean, I pretty much believe none of the confidence intervals.
Everybody is playing very fast and loose with stats.