2026-01-06 05:07 UTC
@trademark@fosstodon.org @HalvarFlake@mastodon.social Some ways I would be wrong:
- If the ai-2027 crowd is right, humans won't have to understand the software. This is an argument for (at least) localized superintelligence, basically.
- If the LLM can construct abstraction layers so well designed and reliable that devs don't have to understand them, (like the OS usually is for applications), humans can spend their time on other things. But, I suspect this is basically superintelligence too.
Replies (1)
-
@trademark@fosstodon.org 2026-01-07 12:15
@headmold@mastodon.social @HalvarFlake@mastodon.social Hmm, you don't seem to understand the argument so let me try again. I have always used to write additional code to help me manage/understand the code I'm actually working on. This code does not need to be as good/resilient/secure as the "real" code. Quite often it just needs to work once. This is a good fit for the LLM, I can just ask for something, continue what I was doing, and come back later instead of repeatedly switching gears and losing state.