The following is written allegedly by Stephen Rice according to the author of the group but it's not confirmed by anyone else. Anyway there will be no confirmation.
Briefly: I don't consider either Loglan or Lojban viable auxlangs.
They weren't designed for it. Their use of logical predicate
structure, while making simple sentences easy to produce, also bloats
the lexicon, because you technically need a new predicate every time
you change the underlying structure--something regular languages use
adpositions to do. Unfortunately, they weren't designed for ease of
derivation, either: Loglanists were originally supposed to chain
together individual predicate words, much as in Toki Pona. The
language was designed for that--it still is, despite some retrofits.
But it doesn't work. At least it doesn't if your predicate words are
all [CCV/CVC]CV in form and you have no good way to borrow. (Loglan
was originally supposed to be an experimental language, remember, not
necessarily a full-fledged everyday language.) So soon groda madzo
("big make"/enlarge) became groma, clika rando sonma madzo ("similar
end sound make"/rhyme) became cliransonma, and so on--without any real
derivational system.
Now, this wasn't entirely a bad thing, because there was no system for
figuring out the structure of these new predicates either, and if a
predicate's structure is incomplete, you begin generating nonsense
sentences the moment you do much with negation or quantification.
Lojbanists ignore this problem for some reason, though it's fairly
obvious, with the result that Lojban is pretty much designed to
produce gibberish for at least some predicates.
Anyway--to simplify derivation, the Great Morphological Revision
modified permissible predicate shapes and introduced
"djifoa"--abbreviated "affixes" for at least the more productive
roots. You can't predict which roots will have djifoa, nor what their
form will be, but once you memorize the djifoa, formal derivation is
fairly automatic.
This is clearly an awkward solution--a retrofit to fix a fatal design
flaw. What still astonishes me is that when the Lojbanists left
Loglan, they faithfully copied both flaw and retrofit. I would've
redesigned the system completely to eliminate the problem from the
beginning, but when I mentioned this to various Lojbanists, including
LeChevalier, their leader, they all replied that the djifoa system was
ingenious and worth preserving--they even took credit for it.
I've considered releasing a total reboot of Loglan without the
morphological and derivational bugs, but I think languages based on
logical predicates are inherently unnatural. Human languages are based
on linguistic predicates, which are in turn based on an overarching
web of relationships (case, etc.), not on mere place structure. One of
my personal oddities as a Loglanist was that I actually memorized the
place structures of predicates; most Loglanists (and Lojbanists, I
think) just memorized the basic meanings. But there's a difference
between
madzo = make
and
madzo = X makes Y from/out of Z.
Compare this with
brudi = X is a brother of Y through parents Z.
From a linguistic standpoint, the relationships differ sharply; as
logical predicates, however, their structures are identical. Loglan
mitigated this with a system of case tags, which the Lojbanists
rejected. My Loglan 2.0 would be based on such a case system.