1/
I once got a ~10x speedup on an external API I had zero access to.
No vendor escalation.
No architectural rewrite.
No concurrency tricks.
Just questioning what the REST layer was hiding.
🧵
Unix’s elegant pipe operator (|)—allowing sysadmins to effortlessly chain commands together—was invented largely because of programmer laziness. Douglas McIlroy, fed up with writing tedious intermediate files and shuffling data manually, demanded a simpler solution. His colleagues at Bell Labs grudgingly obliged, introducing the pipe as an inter-process communication mechanism. Laziness triumphed spectacularly: pipes not only saved effort, but also became a cornerstone of Unix’s elegant modular design philosophy. Decades later, we have sysadmins piping commands together into monstrous, unmaintainable one-liners—exactly as McIlroy intended.
@zilm13 А ещё чё вспомнил: имена чатов ГПТ даёт такие, что список чатов очень сложно использовать. Думаю, им надо после небольшого накопления контекста его переименовывать.
@zilm13 ну да, у меня два и, в общем-то, всё нормально, я могу или вспомнить, где это было или восстановить логику, почему я бы пошел именно туда.
поиск в гпт ещё и медленный, и, вообще, весь интерфейс оставляет желать лучшего, особенно в длинных чатах
Settlers II (1996) supports split-screen multiplayer with two mice.
One must be a serial mouse.
I wrote a tiny Python adapter converting Linux mouse events into the MS serial mouse protocol so DOSBox could see a second mouse.
https://t.co/USDIgdIr3K
#linux#retrocomputing
5/
The important part wasn’t the speedup.
It was recognizing that the abstraction (REST + pagination) was hiding a cost model.
Sometimes performance problems aren’t in your code.
They’re in the layer below the interface.
Full write-up:
https://t.co/PAFs9xBjiu
1/
I once got a ~10x speedup on an external API I had zero access to.
No vendor escalation.
No architectural rewrite.
No concurrency tricks.
Just questioning what the REST layer was hiding.
🧵
4/
Instead of one broad query, I fetched results day by day.
Smaller slices → better index selectivity → tiny offsets → less wasted work.
Same API.
Different mental model.
~10x faster.