If you know me (and you do) you know I'm thinking about Alan Kay's talk
on systems systems last year, and retrospectively Sussman's related talk
from 2011 mhcat shared with me.
One gist was that we got really good at 80s style programming by the
80s, and everything similar since has been spinning wheels. You and I
know this strikes a chord with me- having surveilled the moderne I found
the modern and it was Common Lisp The Language 2e from about 1990, and
the automated theorem common lisp applicative subset ACL2. Let's not
fight. This was as good as a programming language got, and it got and
stayed there approaching four decades now. Common Lisp's Kaiju stability
outstrips changy languages like ANSI C/C++, rnrs schemes and
one-compiler basically-proprietary languages. Sussman hails Haskell as
the most advanced defunct language, while praising Fortran for its
enduring and forward-looking technical merit.
Kay and Sussman point out that Kay's Computing Revolution Hasn't
Happened Yet, and throw more breadcrumbs ahead of us for us to follow.
While they spoke about how rewriting 80s programs won't get us anywhere,
I think the elephant-in-the-room failure is dockerisation and devops
container orchestration generally.
Re-re-writing and running the same 80s programs inside chroot(1) (1990)
was not a backboardless closet to another world despite having been
productised as such for a few decades now (well, semi-proprietary layers
on top of chroot(1)). When was the last time you actually used
jailkit(8) for that matter?
We: You, myself, Kay, Sussman don't actually know where this new
dimension is hiding, which makes this game of hide and seek embarrassing
for others to watch.
I think we know it's not the places we have already looked and told
others we found it: It's not just chroot(1) (1990), not coroutines, it's
not just backpropagation (1975), not proprietary GPT2 general language
model derivatives, it's not diffusion model
text2memorised-copyrighted-pic/GPLed-code. It's not message passing,
it's not job queuing, and streaming multiprocessors (*PUs) are just as
intrinsically unrevolutionary in and of themselves as CPUs.
Throwing my two bits-
One, I think igorstw is right and it's something about AI that
megacorporate players are intrinsically blind to, maybe a truth only AIs
know that we haven't earned their trust sufficiently to be let in on -
hinted at by first order logical proofs only AIs can find.
Two, in hindsight it will turn out to have been on the tips of the
tongues of 80s AI lispers most blamed for bringing this harsh winter
curse to our lands.
ADDENDUM:
I see igorstw left me a message telling me to look at the redox-os new
operating system- microkernel architecture and inspired by plan9.
