Lost in Diffusion: Uncovering Hallucination Patterns and Failure Modes in Diffusion Large Language Models
For researchers and practitioners using diffusion LLMs, this work highlights a critical reliability issue that must be addressed for safe deployment.
This paper presents the first controlled study comparing hallucination in diffusion LLMs vs. autoregressive models, finding that dLLMs hallucinate more. It identifies unique failure modes like premature termination and incomplete denoising.
While Diffusion Large Language Models (dLLMs) have emerged as a promising non-autoregressive paradigm comparable to autoregressive (AR) models, their faithfulness, specifically regarding hallucination, remains largely underexplored. To bridge this gap, we present the first controlled comparative study to evaluate hallucination patterns in dLLMs. Our results demonstrate that current dLLMs exhibit a higher propensity for hallucination than AR counterparts controlled for architecture, scale, and pre-training weights. Furthermore, an analysis of inference-time compute reveals divergent dynamics: while quasi-autoregressive generation suffers from early saturation, non-sequential decoding unlocks potential for continuous refinement. Finally, we identify distinct failure modes unique to the diffusion process, including premature termination, incomplete denoising, and context intrusion. Our findings underscore that although dLLMs have narrowed the performance gap on general tasks, their distinct hallucination mechanisms pose a critical challenge to model reliability. Our code is available at https://github.com/ZeroLoss-Lab/Lost-in-Diffusion