The field of language generation and audio understanding is currently moving towards exploring alternative models to traditional autoregressive approaches. Diffusion-based language models have emerged as a promising alternative, offering improved controllability and robust generation. These models are being applied to various domains, including audio understanding and speech conversation. Researchers are also investigating the use of feedback and noise in language generation, as well as the development of more efficient decoding algorithms. Notable papers in this area include Diffusion Beats Autoregressive in Data-Constrained Settings, which demonstrates the advantage of diffusion models in data-scarce settings, and DIFFA, which introduces a diffusion-based Large Audio-Language Model for spoken language understanding. Additionally, the paper Wide-In, Narrow-Out: Revokable Decoding for Efficient and Effective DLLMs presents a novel decoding algorithm that improves the quality-speed trade-off in Diffusion Large Language Models.