aiqbench analysis
DynHD addresses hallucination detection in diffusion large language models by leveraging denoising dynamics deviation rather than relying solely on static token-level uncertainty. The method improves reliability by filtering out task-invariant structural tokens and measuring discrepancies between observed and expected uncertainty trajectories during the generation process. This approach offers a more efficient and accurate solution for validating factual consistency in next-generation generative models.
DynHD: Hallucination Detection for Diffusion Large Language Models via Denoising Dynamics Deviation Learning
arXiv:2603.16459v2 Announce Type: replace Abstract: Diffusion large language models (D-LLMs) have emerged as a promising alternative to auto-regressive models due to their iterative refinement capabilities. However, hallucinations remain a critical issue that hinders their reliability. To detect hallucination responses from model outputs, token-level uncertainty, such as entropy, has been widely used to indicate potential factual errors. Nevertheless, unlike auto-regressive models that generate tokens sequentially, D-LLMs generate fixed-length sequences simultaneously, where only a small subset of tokens is informative for hallucination detection. Thus, aggregating uncertainty over all tokens can be suboptimal. Moreover, the evolution trend of uncertainty throughout the diffusion process can also provide valuable signals, highlighting the necessity of modeling its denoising dynamics for hallucination detection. In this paper, we propose DynHD, which bridges these gaps from both spatial (token sequence) and temporal (denoising dynamics) perspectives. To handle the information density imbalance across tokens, we propose a semantic-aware evidence construction module that extracts hallucination-indicative signals by removing task-invariant structural tokens and emphasizing the uncertainty of the remaining informative tokens. To model denoising dynamics for hallucination detection, we introduce a reference evidence generator that learns the expected evolution trajectory of uncertainty evidence, along with a deviation-based hallucination detector that makes predictions by measuring the discrepancy between the observed and reference trajectories. Extensive experiments demonstrate that DynHD consistently outperforms state-of-the-art baselines while achieving higher efficiency across multiple benchmarks and backbone models. The code is available at: https://github.com/qyy11-com/DynHD.
Abstract:Diffusion large language models (D-LLMs) have emerged as a promising alternative to auto-regressive models due to their iterative refinement capabilities. However, hallucinations remain a critical issue that hinders their reliability. To detect hallucination responses from model outputs, token-level uncertainty, such as entropy, has been widely used to indicate potential factual errors. Nevertheless, unlike auto-regressive models that generate tokens sequentially, D-LLMs generate fixed-length sequences simultaneously, where only a small subset of tokens is informative for hallucination detection. Thus, aggregating uncertainty over all tokens can be suboptimal. Moreover, the evolution trend of uncertainty throughout the diffusion process can also provide valuable signals, highlighting the necessity of modeling its denoising dynamics for hallucination detection. In this paper, we propose DynHD, which bridges these gaps from both spatial (token sequence) and temporal (denoising dynamics) perspectives. To handle the information density imbalance across tokens, we propose a semantic-aware evidence construction module that extracts hallucination-indicative signals by removing task-invariant structural tokens and emphasizing the uncertainty of the remaining informative tokens. To model denoising dynamics for hallucination detection, we introduce a reference evidence generator that learns the expected evolution trajectory of uncertainty evidence, along with a deviation-based hallucination detector that makes predictions by measuring the discrepancy between the observed and reference trajectories. Extensive experiments demonstrate that DynHD consistently outperforms state-of-the-art baselines while achieving higher efficiency across multiple benchmarks and backbone models. The code is available at: this https URL.
Accepted by EMNLP 2026 Findings. 16 pages, 8 figures, 5 tables
Computation and Language (cs.CL)
arXiv:2603.16459 [cs.CL]
(or arXiv:2603.16459v2 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2603.16459 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 336 words
Related coverage
DeMTS: Denoising Trajectories as Multivariate Time Series for Hallucination Detection in Diffusion Language Models
arXiv · Aug 18, 2026
Latent reasoning architectures would likely undermine CoT, our strongest oversight tool
AI Alignment Forum · Sep 23, 2026
Best Open Speech Recognition (ASR) Models in 2026: WER, Languages, Latency, and License Compared
MarkTechPost · Jul 23, 2026
HIVE: Hidden-Evidence Verification for Hallucination Detection in Diffusion Large Language Models
arXiv · Sep 24, 2026