Education
- Ph.D. in Artificial Intelligence, University of Milano-Bicocca, 2023–2028 (expected: March 2028)
- Curriculum: Big Data & Analytics for Business · Supervisor: Fabio Mercorio
- Visiting PhD student, Nanyang Technological University (NTU) Singapore, May–Dec 2026 (host: Erik Cambria)
- M.Sc. in Data Science, University of Milano-Bicocca, 2021–2023
- B.Sc. in Computer Science, University of Milano-Bicocca, 2018–2021
Research interests
- Mechanistic interpretability: activation steering, role vectors, sparse autoencoders
- Evaluation of LLMs and of interpretability methods
- Deception, fragility and safety of LLMs
Experience
- 2023–present: PhD researcher, University of Milano-Bicocca (CRISP research centre)
- 2019–2023: Freelance software developer (Go, Java, C++, Python)
Skills
- Python, PyTorch, Hugging Face; large-scale experiments on HPC (Leonardo, CINECA/EuroHPC)
- Go, Java, C++
- Languages: Italian (native), English
Publications
- Diagnosing LLM Fragility: An Empirical Study on Reasoning Entropy and Malicious Outputs
Daniele Potertì, et al. · Proceedings of EMNLP 2026 (Main Conference), 2026 - Challenging the Abilities of Large Language Models in Italian: a Community Initiative
Malvina Nissim, Danilo Croce, Viviana Patti, et al. (incl. Daniele Potertì) · Italian Journal of Computational Linguistics (IJCoL), 2026 - Can Role Vectors Affect LLM Behaviour?
Daniele Potertì, Andrea Seveso, Fabio Mercorio · Findings of the Association for Computational Linguistics: EMNLP 2025, 2025 - SFAL: Semantic-Functional Alignment Scores for Distributional Evaluation of Auto-Interpretability in Sparse Autoencoders
Fabio Mercorio, Filippo Pallucchini, Daniele Potertì, Antonio Serino, Andrea Seveso · Proceedings of EMNLP 2025: Industry Track, 2025
Authors in alphabetical order; I contributed to the technical work. - A Benchmark to Evaluate LLMs' Proficiency on Italian Student Competencies
Fabio Mercorio, Mario Mezzanzanica, Daniele Potertì, Antonio Serino, Andrea Seveso · ECML PKDD 2025, 2025
Authors in alphabetical order; I contributed to the technical work. - ITALIC: An Italian Culture-Aware Natural Language Benchmark
Andrea Seveso, Daniele Potertì, Edoardo Federici, Mario Mezzanzanica, Fabio Mercorio · Proceedings of NAACL 2025 (Long Papers), 2025 - Designing Role Vectors to Improve LLM Inference Behaviour
Daniele Potertì, Andrea Seveso, Fabio Mercorio · arXiv preprint, 2025 - BEEP - BEst DrivEr’s License Performer: A CALAMITA Challenge
Fabio Mercorio, Daniele Potertì, Antonio Serino, Andrea Seveso · Proceedings of CLiC-it 2024, 2024
Authors in alphabetical order; I contributed to the technical work. - Augmenting XAI with LLMs: A Case Study in Banking Marketing Recommendation
Alessandro Castelnovo, Roberto Depalmas, Fabio Mercorio, Nicolò Mombelli, Daniele Potertì, Antonio Serino, Andrea Seveso, Salvatore Sorrentino, Laura Viola · World Conference on Explainable Artificial Intelligence (xAI 2024), 2024
Authors in alphabetical order; I contributed to the technical work. - Disce aut Deficere: Evaluating LLMs Proficiency on the INVALSI Italian Benchmark
Fabio Mercorio, Mario Mezzanzanica, Daniele Potertì, Antonio Serino, Andrea Seveso · arXiv preprint, 2024
Authors in alphabetical order; I contributed to the technical work. - Marrying LLMs with Domain Expert Validation for Causal Graph Generation
Alessandro Castelnovo, Riccardo Crupi, Fabio Mercorio, Mario Mezzanzanica, Daniele Potertì, Daniele Regoli · AIABI Workshop @ AI*IA 2023, 2023
Authors in alphabetical order; I contributed to the technical work. - Bridging Causal Theory and LLMs: A User Centric Approach to Causal Graph Generation
Daniele Potertì · Doctoral Consortium @ AI*IA 2023, 2023