Large Language Models (LLMs) are receiving much attention since the recent breakthroughs in text generation. Despite their significant advances, these models often produce counterfactual results, which undermines their reliability and usability. Further, their low interpretability impedes the integration of LLMs into real-world applications. Know-AI focuses on information extraction (IE) for monitoring of global media. The complex hierarchical structure of IE requires the correct analysis of lexical, syntactic, semantic and discourse structures, as well as ability for reasoning, generalization, and accessing real-world knowledge. This makes IE an ideal ground for studying how LLMs capture linguistic phenomena, whether they can encode real-world knowledge, and if so, how. On the practical side, we study how we can adapt LLMs to improve the performance on complex IE tasks. On the theoretical side, we plan to advance the reliability, interpretability and sustainability of LLMs.

2024

2026