Engineering
TLive-Omni: An Omni-Modal Understanding Model for E-Commerce Live Streaming
Researchers developed TLive-Omni, a multimodal model that processes image, video, audio, and text inputs into a unified representation space for…
Engineering
Unlocking the Potential of Image Editing via Concept Scaling and Dense Supervision
Researchers introduce ConceptEdit-12M, a 12-million-pair dataset covering more than 1,000 fine-grained edit concepts to improve image editing models, alongside a…
Engineering
Block3D: Efficient Text-to-3D Generation via Block-Wise Diffusion
Researchers introduce Block3D, a text-to-3D generation framework that partitions discrete shape-token sequences into blocks, generates them sequentially, and jointly denoises…
Engineering
Towards a Densing Law for User Representation Learning at Billion-Scale Capacity
Ant Group researchers propose the User Behavioral Densing Law, which quantifies how tokenization capacity should scale with data size for…
Engineering
RISE: Adaptive Imagination for World Action Models
Researchers present RISE, an adaptive framework for world action models that dynamically decides when to stop imagination rollouts based on…
Engineering
ReWorld: An Interactive World Model with Long-Horizon Memory
Researchers present ReWorld, an interactive world model that addresses the competing demands of real-time control and long-horizon memory through architectural…
Engineering
GameXpert-Bench: How Far Are Coding Agents from Expert Game Development?
Researchers introduce GameXpert-Bench, an evaluation framework with three benchmark tracks assessing coding agents across the game-development lifecycle: initial creation (GameGen,…
Engineering
ARC: Fair Relative Advantage Comparison in Open-Ended Real-World Interaction
Researchers introduce ARC (Advantage Regularization via Conditioning), a training methodology addressing fairness issues when comparing diverse interaction behaviors in reinforcement…
Engineering
AutoResearch: Insight In, Hallucination Out
Researchers present AutoResearch, a two-stage autonomous research system combining idea generation with idea execution. In the generation phase, the system…
Engineering
Beyond Imitation: Filtering On-Policy Distillation by Reasoning Progress
Researchers present R2-OPD, a method addressing misalignment between teacher-derived rewards and actual reasoning progress in on-policy distillation for language models.…
Engineering
LongWoF-Bench: Evaluating Verifiable Long-Workflow Tasks
Researchers introduce LongWoF-Bench, a benchmark of 778 machine-verifiable tasks spanning code generation, agent synthesis, mathematical reasoning, and rule-following, alongside EvoMap,…
Engineering
What AstroPT Knows About Galaxies, and What That Can Teach Us About LLMs
Researchers developed AstroPT, a transformer trained on millions of galaxy images, as a controlled testbed for interpretability research. By probing…