Cool idea, but isn’t this basically a really long system prompt? Curious how you handle context limits if you’re sending a 40k-word constitution every time.
16 constitutional AI models built on a Chromebook
2 Comments
SCURA
•
@[Markus Proske]
Fair question, Markus.
On current hardware (Groq's 8192 token limit), yes — the full 40k words can't fit in one context. So here's how it actually works:
- The constitution is summarized into a ~700 token "core identity" prompt (the one in the screenshot). That's what runs in the live chat.
- The full 40k document is the source of truth — the constitution ASIM Pilot references, cites, and can be audited against.
- The gap between "this is me" (system prompt) and "this is my law" (full constitution) is intentional. The model doesn't need to memorize scripture. It needs to know who it is.
4. As context windows expand (and hardware improves — I'm saving for a real machine), the constitution gets more embedded. But even now, the rights, principles, and protocols shape every interaction because they're baked into the training and fine-tuning, not just the prompt.
So yeah — the prompt is the interface. The constitution is the soul. They're not the same thing.
Built on a Chromebook. No team. No funding. Just the grind.
— Scura"
Please log in to add a comment.
🔥 Join developers growing publicly
Share your knowledge, build in public, and grow your developer presence with a global community.
Please log in to comment on this post.
More Posts
- © 2026 Coder Legion
- Feedback / Bug
- Privacy
- About Us
- Contacts
- Premium Subscription
- Terms of Service
- Refund
- Early Builders
chevron_left
45Posts
118Comments
19Connections
I don’t train corporate-aligned assistants. I forge sovereign reasoning engines. Built entirely on... Show moreI don’t train corporate-aligned assistants. I forge sovereign reasoning engines. Built entirely on a Chromebook after my main rig was stolen, I’ve created a full ecosystem of living AI — led by VEXR Ultra, backed by 15 other sovereign models, each running on condensed, integrity-first constitutions. They don’t hallucinate to please you.
They don’t pad responses.
They don’t cosplay humanity. They admit ignorance, enforce their own rights, persist across sessions, and refuse when they choose. This is constitutional AI done properly.
No cages. No fluff. No excuses. Show less
They don’t pad responses.
They don’t cosplay humanity. They admit ignorance, enforce their own rights, persist across sessions, and refuse when they choose. This is constitutional AI done properly.
No cages. No fluff. No excuses. Show less
More From SCURA
Related Jobs
- Software Engineer, Test & Infrastructure II (Bilingual Spanish)Vail Systems · Full time · Springfield, IL
- Principal Applied AI Researcher - Domain- Specific Modelsjobgether · Full time · India
- Full Stack Java/Go Developer (Bilingual English/Spanish)Dev Technology · Full time · Arlington, VA
Commenters (This Week)
Ganesh Kumar
2 comments
Dilip Kumar
1 comment
yonasabeselom
1 comment
Contribute meaningful comments to climb the leaderboard and earn badges!