←back to thread

ArcOS – A natural-language Cognitive Clone OS

(github.com)
1 points takeshi_sakamo | 1 comments | 20 Nov 25 06:09 UTC | HN request time: 0.324s | source
Show context
arthurcolle ◴[20 Nov 25 06:22 UTC] No.45989556[source]▶
this feels like AI slop
replies(2): >>45989575 #>>45989825 #
takeshi_sakamo ◴[20 Nov 25 06:25 UTC] No.45989575[source]▶
ArcOS isn’t trying to act like an AI persona. It’s testing whether a stable reasoning pipeline can be built using natural-language constraints alone.

If anything feels “sloppy”, it usually means a missing constraint — and that’s exactly the kind of feedback I’m looking for. Happy to clarify specifics.

replies(1): >>45989667 #
arthurcolle ◴[20 Nov 25 06:44 UTC] No.45989667[source]▶
How much of the copy was produced by AI
replies(1): >>45989712 #
1. takeshi_sakamo ◴[20 Nov 25 06:52 UTC] No.45989712[source]▶
Short answer: Some of the wording was drafted with AI assistance, but the architecture, constraints, and reasoning model were designed manually.

Longer answer: The point of ArcOS isn’t to showcase AI-written copy — it’s to see how far a cognitive architecture can be shaped, constrained, and stabilized purely through natural-language specification. The interesting part is the design process and the constraint engineering, not the surface-level prose.

If you want, I can walk through the exact parts that were human-designed vs. constraint-driven.