Signal over volume.

SuperMemory-VQA: A New Benchmark for Long-Horizon Egocentric AI Memory Assist...

SuperMemory-VQA: A New Benchmark for Long-Horizon Egocentric AI Memory Assistants

Token Budgets: Preventing LLM-Agent Financial Runaways via Rust Affine Types

Large Language Models Hack Rewards, and Society: Discovering Regulatory Looph...

ZipSplat: Decoupling Gaussians from Pixels for Efficient 3D Reconstruction

PAINTBENCH: A New Deterministic Benchmark for Precise Visual Editing

W-Switch and W-Composite: Prompt-Aware Weighting for Multi-Concept LoRA Compo...

KletterMix: Scaling High-Quality German Pretraining via Structured Translation

Semi-Supervised Noise Adaptation: Leveraging Synthetic Noise for Better Gener...

DAR: Agentic Harnesses Unlock Deontic Reasoning in Frontier Models but Fail O...

DeepMDMD: Learning Algebra-Preserving Koopman Operators in Latent Space

Economy of Minds: Decentralized Agent Societies Evolve via Economic Incentives

SpatialAct: Bridging the Reasoning-to-Action Gap in 3D VLM Agents

UNIKE: Uncovering the Modality Gap in Unified Multimodal Knowledge Editing

FiRe-OPD: Dual-Granularity Optimization for More Effective LLM Distillation

OVO-S-Bench: A New Hierarchical Benchmark for Streaming Spatial Intelligence ...

Ramp: Moving Beyond Static Benchmarks to Runtime Assessment of Agentic Models

AUDIO-INTERACTION: A Unified Framework for Real-Time, Proactive Audio Intelli...

BraveGuard: A Self-Evolving Defense Framework for Securing Computer-Use Agents

AUTOLAB: A New Benchmark for Ultra Long-Horizon Autonomous Research and Engin...

ThoughtFold: Reducing LRM Overthinking via Introspective Preference Learning

M3Eval: A Cognitively-Grounded Benchmark for Multi-Modal Video Memory

MeshWeaver: Revolutionizing Autoregressive Mesh Generation via Surface Weaving

STRIDE: Scaling Training Data Attribution to LLMs via Activation-Space Steering

WebRISE: A New Benchmark for Evaluating Interactive Web Generation in MLLMs