Latest posts

  1. Diffusion Language Models Might Have the Right Shape for Agents

    Diffusion language models could give agent harnesses a useful interface for fast drafting, joint decisions, structured infilling, and continuously updated working memory.

  2. Reinforcement Learning for Diffusion Language Models

    What is the policy when text is generated by denoising? A technical guide to the objectives, likelihood estimators, failure modes, and emerging design principles behind RL for dLLMs.

  3. β-OPSD: Deriving with Policy Optimization, Training with Self-Distillation

    A principled way to make on-policy self-distillation smoother, more stable, and more effective for reasoning language models.