AI Coding Live News

Three new studies tackle verification, evaluation, and RL alignment for rule-governed generative world models: OmniVerifier-C1 (arXiv:2610.22355, 32 upvotes) uses Decoupled RL to train a decomposed graded visual verifier that reaches 0.684 on ViVerBench and boosts FLUX.1-dev GenEval from 0.68 to 0.79 while enabling Maze and Sudoku world modeling; Meituan's WorldCast benchmark (arXiv:2610.20731, 30 upvotes) evaluates 12 models across 320 multi-step world state prediction cases, finding GPT-Image-2.5 System at 72.8 vs. 42.8 for Nano Banana Pro and 20.1 for the top open model; and NUS's Routing Policy Optimization (RPO, arXiv:2610.22282, 23 upvotes) turns attention routing into an internal exploration action to train video/image diffusion RL 3.1x–4.2x faster than Flow-GRPO and DanceGRPO.

Zoom

Masked diffusion language models (dLMs, such as LLaDA) represent a potent parallel decoding alternative to autoregressive architectures for complex reasoning. However, dLMs face a severe credit-assignment challenge during multi-step denoising: a handful of pivotal commitments sharply reduce uncertainty over remaining masked positions and determine response validity, yet standard post-training schemes naively supervise final text or apply scalar rewards across entire denoising steps. Researchers from KAIST and University of Toronto introduce Pivot-SD (arXiv:2610.03665), an offline self-distillation framework that isolates these pivotal tokens using an information-gain metric. Successful pivots are reinforced via cross-entropy while failure pivots receive targeted unlikelihood penalties—leaving harmless intermediate tokens untouched. Utilizing only 200 prompt questions and four rollouts each (800 rollouts total), Pivot-SD elevates LLaDA-8B-Instruct past both full-sequence SFT and compute-matched diffusion RL baselines across math and coding benchmarks.

Zoom

Shanghai AI Laboratory has open-sourced AV-GRPO (arXiv: 2609.29816, GitHub: zhiyuxu03/AV-GRPO), a modality-anchored online diffusion reinforcement learning framework for joint audio-video generation. Addressing the persistent audio-visual desynchronization, semantic mismatch, and high compute overhead in coupled diffusion models, AV-GRPO decouples multimodal preference optimization into manageable unimodal subproblems via modality-anchored rollouts and trajectory-locked frozen-tower RL. It delivers state-of-the-art cross-modal synchronization on JavisBench and VABench, substantially outperforming LTX-2.3.

Zoom