Meta-Cognitive Analysis: Evaluating Declarative and Procedural Knowledge in Datasets and Large Language Models
This provides primary guidance for evaluating and enhancing large language models, addressing a gap in meta-cognitive analysis for AI researchers.
The paper tackles the lack of comprehensive analysis of declarative and procedural knowledge in large language models by providing ground-truth knowledge and evaluating effective scores, finding that declarative knowledge benefits most tasks more than procedural knowledge, except in simple reasoning tasks, and that model ability improves with pre-training and size at different speeds.
Declarative knowledge and procedural knowledge are two key parts in meta-cognitive theory, and these two hold significant importance in pre-training and inference of LLMs. However, a comprehensive analysis comparing these two types of knowledge is lacking, primarily due to challenges in definition, probing and quantitative assessment. In this paper, we explore from a new perspective by providing ground-truth knowledge for LLMs and evaluating the effective score. Through extensive experiments with widely-used datasets and models, we get conclusions: (1) In most tasks, benefits from declarative knowledge are greater than those from procedural knowledge. (2) Profits of procedural knowledge are larger than declarative knowledge only in reasoning tasks with simple logic. (3) As pre-training progresses and size increases, model ability to utilize both kinds of knowledge significantly improves, but in different speed. We do detailed analysis for the findings and this can provide primary guidance for evaluation and enhancement of large language models.