This paper introduces User-Centric Error Modeling (UCEM), a conceptual framework that redefines personalization in language model systems. Rather than adapting to surface-level preferences or optimizing toward a singular notion of correctness, UCEM proposes that models should learn individualized definitions of error explicit, user-provided explanations of what constitutes an incorrect response relative to specific goals, reasoning patterns, domain assumptions, and working constraints. We argue that meaningful long-term personalization requires models to internalize user-specific error semantics through iterative feedback loops, moving beyond preference-based customization toward what we term error-based cognitive personalization. This paradigm positions users as active co-designers of their model's cognitive boundaries and raises fundamental questions about responsibility, epistemic alignment, and the nature of human-AI collaboration. We present UCEM not as a technical solution but as a theoretical repositioning of the personalization problem, outlining design principles, philosophical implications, scientific challenges, and open research questions necessary to operationalize this vision.
Authors: Momen Ghazouani
Publish Year: 2026
Download PDF