Danielle Ensign
Bio
Danielle Ensign (also known online as Phylliida) is an AI safety researcher and software developer based in the United States. She received a bachelor's degree from the University of Utah (2012–2017), where she published early research on algorithmic fairness, including the influential paper "Runaway Feedback Loops in Predictive Policing" (FAT/ML 2017). She later participated in the OpenAI Scholars program (2021), studying how text generation models affect opinion ecosystems. As a MATS fellow (2024), mentored by Adrià Garriga-Alonso, she conducted circuit-based mechanistic interpretability research on the Mamba state-space model architecture, producing the paper "Investigating the Indirect Object Identification circuit in Mamba" (arXiv:2407.14008) and creating MambaLens, a widely-used TransformerLens port for Mamba. She subsequently became an Anthropic Fellow (2025), working with mentor Kyle Fish on studying LLM conversation termination behavior, resulting in the paper "The LLM Has Left The Chat: Evidence of Bail Preferences in Large Language Models" (arXiv:2509.04781).
Links
- Personal Website
- https://phylliida.dev/
- Twitter / X
- LessWrong
- danielle-ensign
Grants
from Long-Term Future Fund
Discussion
Sign in to join the discussion.
No comments yet. Be the first to share your thoughts.
Details
- Last Updated
- Mar 22, 2026, 3:24 PM UTC
- Created
- Mar 20, 2026, 2:49 AM UTC