⁌ Ethical Training of GaMS LLM (Generative Model for Slovene) (2025-ongoing)

The main objective of this project is to develop and implement a comprehensive, culturally specific, and accountability-centered framework for ethical training of LLMs, with a concrete focus on GaMS. Rather than treating ethical alignment as the instillation of values or seeming moral agency in AI systems (via ‘constitution’), the project approaches ethics training as a normative, epistemic, and institutional practice: one that structures how systems respond to ethically salient situations, how reasons are articulated, and how the bulk of human accountability is preserved.

The project’s core methodological innovation is Supervised Instruction Fine-Tuning with Rationales (SIFT-R), which integrates explicit, culturally grounded ethical justifications to enhance moral intelligibility, transparency, and explainability.

⁌ Epistemic Identity and Epistemic Virtue: Human Mind and Artificial Intelligence (2023-2024)

The project focuses on the concepts of epistemic identity and virtue and interrelate them within the juxtaposition between the human mind and artificial intelligence (AI).

The research comprises of three domains: (i) the notion of epistemic identity and epistemic virtue of human agents, (ii) the aspects of epistemic identity within the interaction between humans and AI and (iii) the ascription of epistemic identity and virtue to AI systems. How does our epistemic identity (i.e. one’s epistemic sensitivity, core beliefs, and ways for assessing knowledge claims) influence epistemic virtues, and how do virtues shape our epistemic identity?

For more, visit the project’s website.

.