Depends
If the number of unique values per sequence is a concern, I'd go for UUIDs. Actually, each sequence can generate 4 billion values (-2147483648 to 2147483647) while the limit per table is 2 billion tuples or 256 gigs. Thus, if you use one sequence per table, usually you won't run out of unique values. If you have more than 2,000 tables (max. number of sequences per database with block size 8192), I'd go for UUIDs too. Sequence values do not get reset if a transaction fails, so they are lost anyway.
Assigning the next sequence value in the create trigger has the advantage, that you can do something with the primary key value immediately after the insert, e.g. assigning it to (new) childs.