Does Out-of-Sight Equal Out-of-Mind in CoT Monitorability?
Pedro Ferreira, Wilker Aziz, Ivan Titov.
🏆 Best Paper (main track). Oral presentation.
Pedro Ferreira
PhD candidate in AI · University of Amsterdam
I am a final-year PhD candidate at the University of Amsterdam (UvA), supervised by prof. Wilker Aziz and prof. Ivan Titov, as part of the Mercury Machine Learning Lab with Booking.com. My main research interests are related to the safety and trustworthiness of language models, in particular the faithfulness and monitorability of their reasoning. More broadly, I am interested in interpretability, post-training methods, and evaluation.
In 2025 I was a research intern at NAVER Labs Europe, working on fine-tuning LLMs to satisfy rare constraints, a project that received an internal Distinguished Award. Before my PhD, I worked in industry building ML systems at Priberam, Unbabel and Farfetch. I obtained an integrated master's degree in Electrical and Computer Engineering from IST, University of Lisbon, with a thesis at Unbabel supervised by prof. André Martins.
I am on the job market for ML research and engineering roles, starting in early 2027.
Email · Google Scholar · GitHub · LinkedIn
Experience
- 2022 – 2026
PhD Candidate, University of Amsterdam (industry-funded by Booking.com)
- 2025
Research Intern, NAVER Labs Europe
- 2020 – 2021
Research Scientist, Priberam
- 2019 – 2020
Junior AI Researcher, Unbabel
- 2018 – 2019
Junior Data Scientist, Farfetch
- 2017 – 2018
Undergraduate Research Intern, Unbabel
Publications
Truthful or Fabricated? Using Causal Attribution to Mitigate Reward Hacking in Explanations
Pedro Ferreira, Wilker Aziz, Ivan Titov.
🏆 Outstanding Paper at the Workshop on Actionable Interpretability, ICML 2025.
Explanation Regularisation through the Lens of Attributions
Pedro Ferreira, Ivan Titov, Wilker Aziz.
Oral presentation.
Priberam Labs at the 3rd Shared Task on SlavNER
Pedro Ferreira, Rúben Cardoso, Afonso Mendes.
🏆 1st place, SlavNER Shared Task.
Sparse and Constrained Attention for Neural Machine Translation
Chaitanya Malaviya, Pedro Ferreira, André F. T. Martins.
Talks
- 2026
Efficient Reasoning in Large Language Models
Internal talk, Booking.com.
- 2025
Reward Hacking in Large Language Models
Internal talk, Booking.com.
- 2023
Slot Attention as a Tool for Interpretability and Better Representation Disentanglement
Booking.com Science and Analytics Conference.
Teaching & Supervision
- 2026
Co-supervisor, MInf thesis, University of Edinburgh
Lyle Buist, Using Linear Probes to Detect and Reduce Unfaithful CoT-Reasoning in LLM Responses.
- 2024
Invited Lecturer, University of Amsterdam
- Language, Speech, and Dialogue Processing (BSc AI)
- 2022 – 2024
Teaching Assistant, University of Amsterdam
- Natural Language Processing I (MSc AI, 2023; 2024)
- Language, Speech, and Dialogue Processing (BSc AI, 2023)
- Basic Probability Theory (MSc Logic, 2022)
- 2019 – 2023
Instructor, Lisbon Machine Learning School (LxMLS)
- Labs team, 2019, 2021 and 2023 editions
- 2018
Instructor, Lisbon Data Science Academy
- Instructor of the 2018 edition, with ~50 students
- Organized the NLP part of the curriculum, together with three other instructors