I ran it on a cluster of 5 of the largest mainframes available at the time, so i don't know what it'll be like stability and footprint-wise on linux/12G memory. I liked the language but i see what people are doing in R, python, julia especially, and numeric libraries in clojure, go, scala, rust, etc they're pretty impressive. One frequent limitation is not too many langaugea have complete wrappers for either of the 2 API's that CUDA exposes, which is pretty important in deep learning.
so if there are tangible advantages to APL today, it's not that obvious.
so if there are tangible advantages to APL today, it's not that obvious.