Intent Recognition in Large Language Models

Prolonged interactions with Large Language Models (LLMs) often result in cognitive degradation resembling semantic drift, contextual confusion, and loss of goal alignment colloquially perceived as "dementia-like" behaviour. This paper introduces a lightweight, prompt-based architectural protocol called Intent Recognition Protocol (IRP) that anchors model responses in user intent rather than superficial dialogue history. The IRP is designed to operate as a top-layer directive within multi-layer artificial consciousness systems.