Filip Sondej

Filip Sondej is a Researcher at the Department of Cognitive Science within the Faculty of Philosophy at Jagiellonian University in Krakow, Poland. His work bridges cognitive science and artificial intelligence, focusing on critical safety aspects of modern language models and multi-agent systems.

His primary research interests include AI safety, LLM unlearning techniques, Chain-of-Thought faithfulness, AI conflict resolution, and digital sentience. Sondej's work addresses fundamental challenges in ensuring that increasingly powerful language models behave safely and align with human values.

Analysis of Sondej's publication record reveals a strong interdisciplinary focus combining cognitive neuroscience methodologies with AI safety research. His recent work demonstrates a clear trajectory from traditional cognitive neuroscience investigations toward cutting-edge AI safety mechanisms, particularly in developing methods for removing unsafe behaviors from language models while maintaining functionality. The publications show sophisticated integration of neural network analysis with human cognitive processes.

Sondej collaborates extensively with researchers including Anna Grabowska and Magdalena Senderecka, appearing as co-author on multiple publications in high-impact journals such as NeuroImage, Cerebral Cortex, and Journal of Cognitive Neuroscience. His research program bridges theoretical cognitive science with practical AI safety applications.