Interpretability for LLM Security
Research on understanding and explaining LLM internal states and mechanisms to improve security, detect vulnerabilities, and enable safer steering of model behavior
Interpretability for LLM Security
Research on Large Language Models in Interpretability for LLM Security
Making LLMs Interpretable and Steerable
Using Mutual Information to Explain Sparse Autoencoder Features
Mapping the Uncertainty in LLM Explanations
A novel framework using reasoning topology to quantify explanation reliability
Smarter Data Interpretation via Language Models
A novel method for extracting meaningful features from datasets using LLMs
Engineering Safer AI Representations
A new approach to make LLMs more predictable and controllable
Bridging LLMs and Statistics
A Framework for Statisticians to Understand and Leverage AI Models
Controlling AI Text Generation
Making LLMs safer through causal reasoning in latent space
Human-Centered XAI Evaluation
Using AI-Generated Personas to Assess Explainable AI Systems
Smarter Attacks on AI Systems
How Understanding LLM Internals Creates Better Adversarial Attacks
Mapping the Vulnerable Landscapes of LLMs
Revealing and manipulating adversarial states in language model security
Surgical Privacy for LLMs
Removing PII without compromising performance
Balancing Transparency and Security in AI Reasoning
A policy framework for Chain-of-Thought disclosure in LLMs
