I've noted that dtplyr (released this January 1.0.1) uses as.data.table to bring the variable back to data.table type: https://dtplyr.tidyverse.org/articles/translation.html
I'm a big fan and user of data.table and use it pipeline with dplyr for many years, for which purpose I wrote myself many of those wrapper functions, which are now part of dtplyr.
I'm however using setDT, as I thought it's more efficient as keeps with data.table mentality of assigning by reference.
So I wonder why Hadley is not using it?
And in general - what's more efficient to use of the two, when one needs to convert from data.frame (or tibble) to data.table?