#+TITLE: 026 / 100daystooffload Programming Considered Harmful
#+AUTHOR: screwtape

I have progressed one more article through this anthology. This
book... The commentary and anthology put in clear and culturally
contexted terms so many of the feelings I and our other peers around
the lispy gopher show intuit about and into computer programs. In an
academically citable way no less. While a bit lame, I feel permitted
by this book to embrace the not-yet-successfully eradicated
countercultural riptide of exploratory programming.

The second article develops the idea of domain specific programming
language creation using specifically the host language's own tools,
highlighting a fundamental implication I had kind of missed about
domain specific languages as a programming concept. It's not that the
language is a tool mixed into some vague mishmash: In fact, if
anything is appearing other than your new domain language, this
evidences deficiency, and the language needs more power (or the domain
of interest more constraint).

Further Winograd's article and commentary leading into this suggest
just why loop and format (and do for that matter, and watching Andrew
Kravchuk stream, just the templating notation itself) are so
revolutionary a general approach to computing life. The problem, as
developed by Winograd- in 1973! - is that as I said on the lispy
gopher show, tantalising systems are *not* formed by either

a) a linear sum of homogeneously conveyed utterly isolated individual
fact morsels

or

b) One learning strategy that we hope has magical strongly emergent
properties (shovelling coal, water and human life into LLM scrapers)

but they are formed of another thing: Complex components. Winograd
considers hackers capability of authoring desirable system as the
upper limit of the attainable complexity of the monomers of the larger
system.

In other words there are two levels: Program items the programmer can
program as system pieces, and their assemblage and orchestration as
the system.

LOOP and FORMAT and DO (and I guess DREI and ZWEI and interlisp's DWIM
and CLISP) are domain specific language for forming larger power
components in the first place. So a chunky LOOP or FOR facility makes
for a grander production than an ultra-minimal light-as-a-feather
recursion.

The limit for a program's desirable properties is the size
of component pieces programmers are capable of creating, and this is
what the baseless and strongly negatively correlating pithy advice on
Design Principles destroys, castrating the future of corporate
programming.

Which is kind of good, in the sense that our enemies are dead in the
water. Falling back on shovelling all remaining life and happiness of
the planet into the hell-forges of Feed Forward Network Optimised By A
Second Feed Forward Network LLMs, known-failure strategy (b) out of
desperate denied knowledge of their ownership's failure.