Best demo award for GPTKB at ISWC
News from the research group Knowledge-aware Artificial Intelligence of November 27, 2025
Best demo award for GPTKB at ISWC
The demo showcases GPTKB, a novel system for comprehensively externalizing the factual knowledge of large language models. While LLMs possess substantial internal factual knowledge, analyzing it has typically been limited to small, pre-defined question samples—a practice that introduces a strong availability bias and reveals only a fraction of a model’s actual knowledge (or beliefs).
GPTKB addresses this limitation through recursive querying and systematic consolidation of model responses, enabling, for the first time, a largely complete and constructive materialization of the factual knowledge contained in an LLM.
As a prototype, GPTKB comprises a knowledge base with 101 million triples covering over 2.9 million entities, extracted from GPT-4.1. It allows simultaneous analysis of a model’s knowledge scale, accuracy, biases, cutoff, and consistency, establishing a new foundation for research on LLM knowledge. The system is publicly available at gptkb.org.
Congratulations to the GPTKB team!
