It appears then it can "autocomplete" a racist message - but i presume because all of these similar texts exist on reddit and were fed as training input. So it

It appears then it can "autocomplete" a racist message - but i presume because all of these similar texts exist on reddit and were fed as training input. So it creates a no-nonsense body of text similar to those. Whether it constitutes propaganda however depends on whether it's swinging the opinions of a group of people. Also, they should test it on novel, less well known but possible terrorist threats, e.g. Catalan extremists.