Tag
The paper introduces Physis-Lang, a self-evolving framework that uses physical language as a shared, optimizable representation across data curation, training, and video generation, alongside the PhysCapBench benchmark and an agentic loop for refining physical captions. With open-source Cosmos-Nano backbones, the enhanced models surpass proprietary Veo3.1 on physical plausibility benchmarks.
This paper proposes representing multimodal data (image, video, text) as bags of atomic propositions (e.g., 'person holding cup'), unified via a global semantic codebook, enabling interpretable, compositional, and cross-modal understanding. The framework is demonstrated on autonomous driving and open-world data.
The paper proposes a hybrid pre-training objective combining JEPA latent-space prediction with MLM reconstruction for language models, showing improved embedding uniformity and semantic-lexical balance.
This paper argues that designing advanced language representations to shape cognitive schemas is a key frontier for expanding LLM intelligence without scaling parameters. It provides formalizations and empirical evidence showing that different linguistic structures significantly impact model performance and internal feature activations.