Papers1111

The Trinity of Consistency as a Defining Principle for General World Models

Jingxuan Wei, Siyuan Li et al.Feb 26arXiv

The paper argues that to build an AI that truly understands and simulates the real world, it must be consistent in three ways at once: across different senses (modal), across 3D space (spatial), and across time (temporal).

#world model#trinity of consistency#modal consistency

SkyReels-V4: Multi-modal Video-Audio Generation, Inpainting and Editing model

Intermediate

Guibin Chen, Dixuan Lin et al.Feb 25arXiv

SkyReels-V4 is a single, unified model that makes videos and matching sounds together, while also letting you fix or change parts of a video.

#multimodal diffusion transformer#video-audio generation#inpainting

Test-Time Training with KV Binding Is Secretly Linear Attention

Intermediate

Junchen Liu, Sven Elflein et al.Feb 24arXiv

The paper shows that Test-Time Training (TTT) with key–value (KV) binding is not really memorizing like a notebook; it is acting like a learned linear attention layer.

#Test-Time Training#KV Binding#Linear Attention

Multi-Vector Index Compression in Any Modality

Beginner

Hanxiang Qin, Alexander Martin et al.Feb 24arXiv

Searching through videos, images, and long documents is powerful but gets very expensive when every tiny piece is stored separately.

#multi-vector retrieval#late interaction#index compression

On Data Engineering for Scaling LLM Terminal Capabilities

Intermediate

Renjie Pi, Grace Lam et al.Feb 24arXiv

This paper shows that you can vastly improve a model’s command-line (terminal) skills by carefully engineering the training data, not just by using a bigger model.

#Terminal-Bench 2.0#terminal agents#synthetic task generation

From Perception to Action: An Interactive Benchmark for Vision Reasoning

Beginner

Yuhao Wu, Maojia Song et al.Feb 24arXiv

The paper introduces CHAIN, a hands-on 3D playground that tests if AI can not only see objects but also plan and act under real physics.

#interactive benchmark#vision-language models#physical reasoning

See and Fix the Flaws: Enabling VLMs and Diffusion Models to Comprehend Visual Artifacts via Agentic Data Synthesis

Intermediate

Jaehyun Park, Minyoung Ahn et al.Feb 24arXiv

Modern image generators can still make strange mistakes like extra fingers or melted faces, and today’s vision-language models (VLMs) often miss them.

#visual artifacts#structural artifacts#diffusion transformer

PyVision-RL: Forging Open Agentic Vision Models via RL

Intermediate

Shitian Zhao, Shaoheng Lin et al.Feb 24arXiv

PyVision-RL teaches vision-language models to act like curious agents that think in multiple steps and use Python tools to inspect images and videos.

#agentic multimodal models#reinforcement learning#dynamic tooling

QuantVLA: Scale-Calibrated Post-Training Quantization for Vision-Language-Action Models

Intermediate

Jingxuan Zhang, Yunta Hsieh et al.Feb 23arXiv

Vision-Language-Action (VLA) robots are powerful but too big and slow for many real-world devices.

#Vision-Language-Action#Post-Training Quantization#Diffusion Transformer

Mobile-O: Unified Multimodal Understanding and Generation on Mobile Device

Intermediate

Abdelrahman Shaker, Ahmed Heakl et al.Feb 23arXiv

Mobile-O is a small but smart AI that can both understand pictures and make new images, and it runs right on your phone.

#Mobile-O#unified multimodal model#on-device AI

A Very Big Video Reasoning Suite

Intermediate

Maijunxian Wang, Ruisi Wang et al.Feb 23arXiv

This paper builds a gigantic library of video puzzles (VBVR) so AI can practice not just making pretty videos, but actually thinking through what happens over time.

#video reasoning#rule-based evaluation#in-domain generalization

ManCAR: Manifold-Constrained Latent Reasoning with Adaptive Test-Time Computation for Sequential Recommendation

Intermediate

Kun Yang, Yuxuan Zhu et al.Feb 23arXiv

ManCAR helps recommendation systems think step by step but keeps their thoughts on realistic paths using a map of how items connect.

#sequential recommendation#latent reasoning#interaction graph

1 2 3 4 5