3 ms·

I used GEOS on the C64, it was, and still is an astounding programming accomplishment.

api 3y ago
It puts into perspective just what a spectacular amount of bloat we have in modern software. A lot of it comes from diversity though. GEOS was a tightly coupled system with one design, API, and programming paradigm. Same goes with classic macOS and other old school desktops. You had no more than two or three languages in use (usually ASM, C, and something like Pascal), one GUI library, one set of runtime libraries, etc. Today your typical desktop hosts apps written in dozens of languages using dozens of graphics libraries and rendering layers. Since different apps use different versions of these things each app often ships the entire stack with it, meaning you may be running three or four versions of a dozen things. To make all that interoperable requires a Byzantine mess of APIs and interfaces and across the whole system. There are tons of shims and hacks and duplication of effort. I remember a silent paradigm shift happening in the 2000s when we abandoned the question of what the “best” language or OS was and instead concluded “we’ll just run everything all at once.” So now we need 16 gigabytes of RAM for a desktop. I wonder sometimes if AI could get us out of this mess. Could future much more powerful coding models do things like reimplement the entire desktop environment in our language of choice at our command? Could they consolidate and refactor code at industrial scale? The thing that makes it impossible to get rid of all this cruft is the cost of doing so, something automation could cut.
- theodric 3y ago
  If we could somehow make AI output deterministic and predictable, maybe. At the moment all it does for me is manage to make a 20GB GPU feel inadequate while every output is a loaded dice roll-- the opposite of what you're considering.
  - api 3y ago
    This is early days.
- speed_spread 3y ago
  AI will mostly let us get into a (much) deeper mess by cheaply generating tons of mediocre code. By the time AI gets around to understand complex code, it'll be to fix it's own generated mess.
- Aerbil313 3y ago
  Codegen AI still operates on bare text. Wait for those transformers which will be aware of language rules, scoping, types etc.
ipaddr 3y ago
I used it as well. It was truly astounding.

api 3y ago

It puts into perspective just what a spectacular amount of bloat we have in modern software. A lot of it comes from diversity though. GEOS was a tightly coupled system with one design, API, and programming paradigm. Same goes with classic macOS and other old school desktops. You had no more than two or three languages in use (usually ASM, C, and something like Pascal), one GUI library, one set of runtime libraries, etc. Today your typical desktop hosts apps written in dozens of languages using dozens of graphics libraries and rendering layers. Since different apps use different versions of these things each app often ships the entire stack with it, meaning you may be running three or four versions of a dozen things. To make all that interoperable requires a Byzantine mess of APIs and interfaces and across the whole system. There are tons of shims and hacks and duplication of effort. I remember a silent paradigm shift happening in the 2000s when we abandoned the question of what the “best” language or OS was and instead concluded “we’ll just run everything all at once.” So now we need 16 gigabytes of RAM for a desktop. I wonder sometimes if AI could get us out of this mess. Could future much more powerful coding models do things like reimplement the entire desktop environment in our language of choice at our command? Could they consolidate and refactor code at industrial scale? The thing that makes it impossible to get rid of all this cruft is the cost of doing so, something automation could cut.

theodric 3y ago
If we could somehow make AI output deterministic and predictable, maybe. At the moment all it does for me is manage to make a 20GB GPU feel inadequate while every output is a loaded dice roll-- the opposite of what you're considering.
- api 3y ago
  This is early days.
speed_spread 3y ago
AI will mostly let us get into a (much) deeper mess by cheaply generating tons of mediocre code. By the time AI gets around to understand complex code, it'll be to fix it's own generated mess.
Aerbil313 3y ago
Codegen AI still operates on bare text. Wait for those transformers which will be aware of language rules, scoping, types etc.

ipaddr 3y ago

I used it as well. It was truly astounding.