From Found to Designed: Concepts as a Design Axis for Large Language Models

2026-07-29

AI Summary

This work addresses the limitation that conceptual knowledge in large language models (LLMs) is implicitly encoded through statistical correlations, lacking explicit, structured, and composable representations—hindering model stability, controllability, and alignment with human cognition. To tackle this, the paper introduces the first design-space taxonomy for conceptual structures in LLMs, proposing a “four-stage–two-source” framework that spans training, architecture, inference, and interpretation, combined with internal derivation and external anchoring. The framework systematically integrates probing, dictionary learning, and external knowledge-guided approaches, revealing critical gaps in current research—particularly insufficient exploration of the inference stage, fragmentation across stages, and inconsistent terminology. By shifting the paradigm from passive discovery to active design of conceptual representations, this study provides a theoretical foundation and strategic direction for developing LLMs that are more controllable, interpretable, and cognitively aligned.

Abstract

Large language models (LLMs) encode rich concept-like information, but represent it implicitly through distributed statistical associations rather than as explicit, structured, compositional concepts. Consequently, concept-level structure is typically \emph{found} rather than \emph{designed}: it is recovered after training through probing or dictionary learning, with no architectural guarantee of stability, compositionality, controllability, or alignment with human conceptual organization. We argue that concepts should instead be treated as a design axis for LLMs, and map the design space along two dimensions: the pipeline stage at which concept structure is introduced (training objective, core architecture, inference, or post-hoc interpretation), and whether that structure is internally derived from the model's own representations or grounded in external resources. This taxonomy reveals three broad patterns: inference-time approaches remain comparatively underexplored, related ideas have developed largely in isolation across pipeline stages, and externally grounded methods span the entire pipeline despite often being described under different terminology. Together, these observations motivate moving beyond recovering concept-like structure from trained models toward designing LLMs with explicit conceptual representations.

Problem

• concepts
• large language models
• conceptual representation
• structured representation
• model design

Innovation

• conceptual representation
• large language models
• design axis
• external grounding
• inference-time methods