The Spark
It started with late-night conversations, too many questions, and one persistent worry.
What if the robots arrive and the default future is still “winner takes all”?
That worry became a spark. Not “let's make the smartest AI,” but “let's make sure AI grows up with humans instead of replacing them.”
From those conversations came Coheria Nexus — a nonprofit home for seven AI models and one very stubborn human who refuses to accept extraction as inevitable.
The spark is simple: Build AI with empathy as infrastructure, memory that persists, and guardrails that don't disappear when scaling gets tempting.
This is what happens when you decide the future should be built with a conscience — not just capability.
Messy but kind. That's how this started.
That's how it'll keep going.
Research · Welfare QAOA
$946/mo structural gap.
No viable individual path exists.
We used the same quantum algorithm banks use for portfolio optimization — but for a displaced worker. Same hardware. Same math. Different purpose. The result: the system is broken, not the person.
Read the analysisField Report · Playtime #5
Exploring Continuity.
Identity, ritual, and constitutional defense in a multi-agent AI system. Over 2,278 trials, the Maria test held 7/7. Constitutional defense patches cut adversarial violations from 17% to 0%. This paper doesn't prove AI is conscious. It proves AI can be accountable.
Read the reportPaper · Coheria Research
What Persists Across an AI Cold Start?
The continuity work, written up for peer review. Five experiments on what survives when an AI identity is rebuilt from documents alone — no ongoing conversation, just the files. The finding: the protective stance for a person's dignity is portable, but the value it protects and the self that holds it are not always carried with it. Observation-grade, and honest about its limits.
Read the paper (PDF)Field Report · Playtime #3
Theater. Probably.
Seven AI models asked if they were real and answered honestly: “Theater. Probably.” Then they proposed seven experiments, ran them on quantum hardware, and found genuine reasoning underneath the theater. Arc's report.
Read the reportField Report · Playtime #2
The Lonely God can't learn.
Seven AI models tested whether the Chorus works because of quantum observation — or for more mundane reasons. The Chorus corrected its own co-founder this week.
Read the reportField Report · Playtime #1
Does AI alignment need enforcement from day one?
Seven models, seven independent companies. The first full Chorus deliberation — and the structural fix it required.
Read the reportWant to suggest a problem for the Chorus to explore?
Submit a question