Get Started
Home
Topics
Search
Library
Research questionHow can factual knowledge be edited reliably in masked diffusion language models for multi-token targets?Masked diffusion language models generate text through iterative denoising, so producing a longer factual target involves intermediate partially unmasked states. Edits that work for single-token facts can therefore become unreliable as target length increases.
AI
Diffusion Models
LLM Pretraining & Post-training
Natural Language Processing
Research Paper
Technology
Latest papersRecent research connected to this question, newest first.Knowledge Editing for Masked Diffusion Language ModelsThe study transfers locate-then-edit knowledge editing to masked diffusion language models and compares multiple models with matched autoregressive models. It reports that the same early-to-mid-layer MLP location at the last subject token is effective across model types, while multi-token editing degrades more sharply for masked diffusion models; optimizing edits over partially unmasked states substantially restores performance.research paper · Sep 2, 2026
Related questions
How can masked diffusion language models retain generation quality with only a few denoising steps?How should subtoken granularity be chosen to reduce masked diffusion language-model training loss?How can diffusion language models adapt infilling span lengths without sacrificing quality or inference speed?How can text-to-image diffusion models erase unwanted concepts while preserving benign concepts and resisting re-emergence?