DeLeaker: Dynamic Inference-Time Reweighting For Semantic Leakage Mitigation in Text-to-Image Models Paper • 2510.15015 • Published Oct 16 • 10
CRISP: Persistent Concept Unlearning via Sparse Autoencoders Paper • 2508.13650 • Published Aug 19 • 15
Follow the Flow: On Information Flow Across Textual Tokens in Text-to-Image Models Paper • 2504.01137 • Published Apr 1 • 21
Padding Tone: A Mechanistic Analysis of Padding Tokens in T2I Models Paper • 2501.06751 • Published Jan 12 • 32
LLMs Know More Than They Show: On the Intrinsic Representation of LLM Hallucinations Paper • 2410.02707 • Published Oct 3, 2024 • 47
