It's practically a benchmark now. Some friends have been specifically training models to count the R's in "strawberry"