Probing the Knowledge Boundary: An Interactive Agentic Framework for Deep Knowledge Extraction
This work addresses the challenge of understanding what knowledge LLMs contain for researchers and practitioners, though it is incremental as it builds on existing probing methods with a more systematic approach.
The paper tackles the problem of systematically probing the knowledge boundaries of Large Language Models (LLMs) by proposing an interactive agentic framework with adaptive exploration policies and a knowledge processing pipeline, resulting in findings such as recursive taxonomy being the most effective strategy, a clear knowledge scaling law, and distinct knowledge profiles across model families.
Large Language Models (LLMs) can be seen as compressed knowledge bases, but it remains unclear what knowledge they truly contain and how far their knowledge boundaries extend. Existing benchmarks are mostly static and provide limited support for systematic knowledge probing. In this paper, we propose an interactive agentic framework to systematically extract and quantify the knowledge of LLMs. Our method includes four adaptive exploration policies to probe knowledge at different granularities. To ensure the quality of extracted knowledge, we introduce a three-stage knowledge processing pipeline that combines vector-based filtering to remove exact duplicates, LLM-based adjudication to resolve ambiguous semantic overlaps, and domain-relevance auditing to retain valid knowledge units. Through extensive experiments, we find that recursive taxonomy is the most effective exploration strategy. We also observe a clear knowledge scaling law, where larger models consistently extract more knowledge. In addition, we identify a Pass@1-versus-Pass@k trade-off: domain-specialized models achieve higher initial accuracy but degrade rapidly, while general-purpose models maintain stable performance during extended extraction. Finally, our results show that differences in training data composition lead to distinct and measurable knowledge profiles across model families.