Engineering
LAION-BVD: A 10-Million-Hour Open Video Dataset for Multimodal Pre-training
Researchers introduce a large-scale open video dataset containing 80 million videos totaling 10 million hours of content for multimodal learning…
Engineering
CAFE: Self-Improving Search Agents Need Co-Evolving Feedback
Tencent Hunyuan researchers introduce CAFE, a framework in which a shared-parameter model alternates between agent and critic roles to build…
Engineering
AgentRoom: Concurrent Multi-Agent Coding in a CRDT-Backed Shared Workspace
Researchers present AgentRoom, a real-time collaborative editing protocol that enables concurrent multi-agent LLM coding through a shared filesystem built on…
Engineering
Automata from Agent Traces: Failure and Next-Step Prediction
Researchers convert agent execution traces into compact finite-state machines containing 7 to 43 states that reconstruct held-out data with 0.997…
Engineering
Length-Adaptive Decoding for Masked Diffusion Machine Translation
Researchers present Entropy-Valley, a training-free technique for selecting target canvas lengths in masked diffusion machine translation that scores candidate lengths…
Engineering
DREAM: An Agentic Control Architecture for Recommender Systems
Alibaba researchers describe DREAM, an agentic control architecture that adds a perception-aware policy layer above existing recommendation pipelines, comprising an…
Engineering
When “Must” Becomes “Maybe”: Constraint Weakening in LLM Agent Workflows
Researchers investigate how intermediate language artifacts in multi-stage LLM agent workflows can inadvertently weaken operational constraints, evaluating 1,296 synthetic episodes…
Engineering
MoTE: Mixture of Task Experts for Multi-Task Video Understanding
Researchers propose MoTE, a decoder architecture that converts large language model feed-forward networks into task-specific experts while keeping the multimodal…
Engineering
TorchMorph: CUDA-Accelerated Morphological Transforms for PyTorch
Kai Zhao presents TorchMorph, a lightweight PyTorch extension implementing 22 morphological operators as fused CUDA kernels that operate directly on…
Engineering
Autonomous Mathematical Discovery in an Open-World Multi-Agent Environment
Researchers describe the Station, an autonomous multi-agent mathematical discovery system where AI agents from different model families collaboratively pursue research…
Engineering
Latent Action as Intention Enables Efficient Future Imagination for World Action Models
Researchers developed LAWA, a world action model architecture that represents future intentions using compact latent actions rather than generating explicit…
Engineering
EchoWM: Open and Enterable Omnimodal World Models
Researchers developed EchoWM, an omnimodal world model that generates synchronized 720p video, environmental sound, music, and speech while responding to…