Liv d'Aliberti | Princeton Computer Science

Room 320C, the Messiest Cubicle

Sherrerd Hall

Princeton, NJ, 08544

I’m Liv d’Aliberti, a Computer Science Ph.D. student at Princeton University. I am advised by Manoel Horta Ribeiro and affiliated with the Humans and Machines (HAM) Lab and Center for Information Technology Policy. I study how AI agents learn behavior during training. My research focuses on reliability, interpretability, and empirically grounded evaluation.

research
- Reinforcement learning
- Machine learning
- Reliable reasoning
current work
- RLVR solution diversity
- Judgemental forecasting
- Agentic misalignment
previous life
- Government R&D
- Privacy
- Autonomy

Before my Ph.D., I worked as a research scientist in the CTO Office Artificial Intelligence & Machine Learning Accelerator at Leidos. I also teach modern Python software development and deep learning with PyTorch as a Lecturer in Computer Science at Johns Hopkins Engineering for Professionals. I earned my M.S. in Applied Mathematics and Statistics at Georgetown University in 2019.

I really like all cats and one dog, my wife’s cavapoo. When I’m offline, I’m usually hiking, backpacking, skiing, or exploring new places.

news

Oct 26, 2026	Can Language Models Forecast Inductively? was accepted as a poster at SocialAgent at NeurIPS 2026.
Oct 08, 2026	Preprint released: Measuring and Mitigating Solution Mode Collapse in RLVR, accepted to MATH-AI at NeurIPS 2026.
Jun 17, 2026	Behavior-Consistent Deep Reinforcement Learning was accepted to AutoRL at RLC 2026.
May 26, 2026	I passed my Princeton C.S. generals exam!
May 20, 2026	Preprint released: Behavior-Consistent Deep Reinforcement Learning.

selected publications

MATH-AI

Measuring and Mitigating Solution Mode Collapse in RLVR

Liv G. d’Aliberti , Marwa Abdulhai , Sofiia Druchyna, and 2 more authors

In The 6th Workshop on Mathematical Reasoning and AI at NeurIPS 2026, Dec 2026

Accepted to MATH-AI at NeurIPS 2026

research explained Abs arXiv Paper Bib

@inproceedings{daliberti2026measuringmitigatingsolutionmodecollapse,
  title = {Measuring and Mitigating Solution Mode Collapse in RLVR},
  author = {d'Aliberti, Liv G. and Abdulhai, Marwa and Druchyna, Sofiia and Henderson, Peter and Horta Ribeiro, Manoel},
  booktitle = {The 6th Workshop on Mathematical Reasoning and AI at NeurIPS 2026},
  year = {2026},
  month = dec,
  eprint = {2610.11064},
  archiveprefix = {arXiv},
  primaryclass = {cs.LG},
  url = {https://arxiv.org/abs/2610.11064},
  paper = {https://arxiv.org/abs/2610.11064},
  note = {Accepted to MATH-AI at NeurIPS 2026},
}

AutoRL

Behavior-Consistent Deep Reinforcement Learning

Marcel Hussing , Liv G. d’Aliberti , Claas Voelcker, and 2 more authors

In Workshop on Automated Reinforcement Learning at RLC, Aug 2026

Accepted

research explained arXiv Post Bib

@inproceedings{hussing2026behaviorconsistentdeepreinforcementlearning,
  title = {Behavior-Consistent Deep Reinforcement Learning},
  author = {Hussing, Marcel and d'Aliberti, Liv G. and Voelcker, Claas and Eysenbach, Benjamin and Eaton, Eric},
  booktitle = {Workshop on Automated Reinforcement Learning at RLC},
  year = {2026},
  month = aug,
  eprint = {2605.21214},
  archiveprefix = {arXiv},
  primaryclass = {cs.LG},
  url = {https://arxiv.org/abs/2605.21214},
  post = {https://x.com/livdaliberti/status/2057876641653670026},
  note = {Accepted},
}

ACL Findings

The Illusion of Insight in Reasoning Models

Liv G. d’Aliberti and Manoel Horta Ribeiro

In Findings of the Association for Computational Linguistics: ACL 2026, Jul 2026

research explained DOI arXiv Post Paper Bib Code Poster

@inproceedings{daliberti2026illusioninsightreasoningmodels,
  title = {The Illusion of Insight in Reasoning Models},
  author = {d'Aliberti, Liv G. and Horta Ribeiro, Manoel},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  month = jul,
  address = {San Diego, California, United States},
  publisher = {Association for Computational Linguistics},
  pages = {924--966},
  doi = {10.18653/v1/2026.findings-acl.47},
  eprint = {2601.00514},
  archiveprefix = {arXiv},
  primaryclass = {cs.AI},
  url = {https://aclanthology.org/2026.findings-acl.47/},
  paper = {https://aclanthology.org/2026.findings-acl.47/},
  post = {https://x.com/manoelribeiro/status/2008261102853857393},
}

IEEE Aerospace

Explainability for Unmanned Aerial Vehicle Control via Multi-Objective Reinforcement Learning

Christian A. Clark , Kevin M. Albarado , Joshua P. Wilson, and 3 more authors

In 2025 IEEE Aerospace Conference, Jul 2025

DOI Paper Bib

@inproceedings{11068676,
  author = {Clark, Christian A. and Albarado, Kevin M. and Wilson, Joshua P. and Bernard, Cedric A. and Scott, Jackson D. and D'Aliberti, Liv},
  booktitle = {2025 IEEE Aerospace Conference},
  title = {Explainability for Unmanned Aerial Vehicle Control via Multi-Objective Reinforcement Learning},
  year = {2025},
  volume = {},
  number = {},
  pages = {1-15},
  keywords = {Training;Command and control systems;Visualization;Reinforcement learning;Production;Autonomous aerial vehicles;Robustness;Fuels;Standards;Optimization},
  doi = {10.1109/AERO63441.2025.11068676},
  paper = {https://ieeexplore.ieee.org/document/11068676},
}