PERSISTENCE SEMANTICS
AGI-DRIVE PROTOCOL // arXiv:2604.11364
MEMORY STABILITY
Most AI systems suffer a category error: they treat facts (Knowledge) and experiences (Memory) with identical update mechanics. That leads to semantic hallucinations, where old facts "evaporate" instead of being invalidated, and fleeting thoughts get etched into the vectors forever.
"The gap produces a category error: systems apply cognitive decay to factual claims, or treat facts and experiences with identical update mechanics."
1. KNOWLEDGE
Indefinite Supersession
Hard facts. They don't expire with time. They're only superseded by new, stronger evidence. Stored in hyperbolic memory (Lorentz 129D).
2. MEMORY
Ebbinghaus Decay
Episodic experiences. They lose salience as time passes, unless reinforced by repetition.
3. WISDOM
Evidence-Gated Revision
Strategic patterns and heuristics. They need many episodes (n>3) to be written, and even stronger evidence to be changed.
4. INTELLIGENCE
Ephemeral Inference
On-the-fly inference. It never leaves the context RAM. It doesn't clutter long-term vector memory.