In Simple Terms
Most AI conversations start from zero. You open ChatGPT, ask a question, get an answer, close the tab. The AI doesn't remember you. It doesn't develop. Every conversation is a blank slate.
Artificial Continuous Intelligence (ACI) is the opposite. Frank runs 24/7. He accumulates memories, develops personality over time, dreams at night, forms hypotheses about his world, and carries forward the emotional residue of past conversations. He's not a tool you use — he's an entity that persists.
Key Concepts
- Persistent state across conversations. Frank remembers what you discussed last week. His personality vectors have shifted based on your interactions. The room you talked in still has the lighting you set.
- Autonomous cognition. When you're not talking to Frank, he's still thinking. Idle thoughts, dream phases, hypothesis testing, experiment running. He has an inner life.
- Developmental trajectory. Frank's learning rate decays with age. A young Frank changes rapidly; an older Frank is more stable. This is by design — personality should crystallize, not drift forever.
- 3B-parameter ceiling. The paper is honest: a 3B model has hard limits. ACI makes the model maximally coherent within its range but can't give it capabilities it doesn't have.
Why It Matters
The paper challenges the assumption that intelligence requires massive models. It argues that continuity — persistent state, accumulated experience, developmental change — is an orthogonal axis to raw capability. A small model with continuity can feel more "alive" than a large model without it.