Research Papers
CARD Brings Cluster-Level Personalization to LLM Text Generation
A new hierarchical framework clusters users by shared writing style, learns group-specific LoRA adapters, and applies personalizat...
Study Maps Jev Decision Model's Rapidly Growing Ecosystem Across 2,170 GitHub Projects
A large-scale analysis of 2,170 public GitHub projects finds Jev, a fast low-cost decision model, is being reused across domains f...
FoMo Uses Diffusion Timesteps to Label Image Quality Without Human Annotators
A new Hugging Face paper proposes FoMo, a fully automated pipeline that derives pointwise perceptual distance labels from the fork...
PISA Brings Log-Linear Block Sparse Attention to Long-Context Language Models
A new Hugging Face paper proposes PISA, a block-sparse attention mechanism that uses pyramid Top-K selection to cut attention comp...
InternW0-Δ: A World Action Model Unifying Predictive Dynamics and Robot Actions
Intern Robotics introduces InternW0-Δ, a World Action Model that fuses visual dynamics, semantics, 4D geometry, and motion priors...
FuseReg Trains Autoencoders to Handle Any Layer Fusion, Cutting gFID by up to 29%
A new method replaces fixed heuristic layer fusion in representation autoencoders with random layer-subset sampling during trainin...
World Action Agent Lets VLMs Pilot Robots Through a Visual Action Workspace
A new Hugging Face paper introduces World Action Agent (WAA), a multi-agent harness that lets general-purpose VLMs control robots...
AV-GRPO: A New Reinforcement Learning Framework for Joint Audio-Video Generation
Hugging Face researchers introduce AV-GRPO, described as the first GRPO framework for joint audio-video generation, alongside the...
ExplorationBench Tests Whether AI Can Truly Explore, Not Just Recall
Hugging Face researchers introduce ExplorationBench, a benchmark built on verifiable "Alien Worlds" whose executable rules let eve...
PUBG Ally: Krafton's Voice-Enabled AI Teammate Plays Alongside Real Players
A new research paper introduces PUBG Ally, an embodied conversational agent that reasons, acts, and speaks as a teammate in PUBG:...
Neural Spectral Capacity Scores Transformers From Specs Alone, No Training Needed
A new NeurIPS 2026 paper introduces Neural Spectral Capacity (NSC), a closed-form scalar that ranks and designs Transformer archit...
Coding Agents Beat Hand-Engineered Planners on Generalized Task and Motion Planning
A new Hugging Face paper finds that coding agents such as Claude Code and Codex can synthesize programs that generalize across tas...