Work

Bio

Current

Research Fellow @ Center on Long-term Risk

Researcher @ Cadenza Labs

Visiting researcher @ Constellation Institute

Red-teamer @ Trajectory Labs

Research Fellow @ Pivotal Research

Research Fellow @ Apart Research

Organising & Research Management @ AI Safety Vietnam

I want to know and meet models. Current interests include functional introspection, preferences, identity, and selfhood.

In worlds where alignment is hard, I also care about reliable monitoring and oversight.

Improving Latent Introspection Elicits LLM Self-Reports — TBD.

Activation Space Monitors Can Amplify LLM Oversight — TBD.

Probing and Steering Evaluation Awareness of Language Models — ICML 2025 Workshops (Technical AI Governance & Actionable Interpretability). First author. Cited in Bengio et al. International AI Safety Reports.
Evaluating Precise Geolocation Inference Capabilities of VLMs — AAAI 2025 Workshop (Datasets and Evaluators of AI Safety). Co-first author.
DarkBench: Benchmarking Dark Patterns in Large Language Models — ICLR 2025 Oral Spotlight. Co-first author.
Werewolf Benchmark — Mikaelson, Nguyen, Liu, Nguyen, Kundu
Identifying Cooperative Personalities through Personality Steering — TAIS 2025. Co-author.