What are you looking for?

独家 | 智谱创始人唐杰发内部信：「GLM 时刻」之后，什么是更重要的事

AI·Zhipu AI founder Tang Jie releases internal letter after 'GLM Moment' success, announcing focus on long-horizon tasks, autonomous agents, and self-evolving AI beyond coding and reasoning.

DeepSearch-World: Self-Distillation for Deep Search Agents in a Verifiable Environment

AI·DeepSearch-World is a self-distillation framework for web search agents that uses self-generated experience for training in verifiable environments.

LongE2V: Long-Horizon Event-based Video Reconstruction, Prediction, and Frame Interpolation with Video Diffusion Models

AI·LongE2V leverages pre-trained video diffusion models to reconstruct, predict, and interpolate event-based videos, achieving high data efficiency and superior perceptual quality through autoregressive unrolling and adaptive context switching.

Remember When It Matters: Proactive Memory Agent for Long-Horizon Agents

AI·Proactive memory agent introduced to combat behavioral state decay in long-horizon agents by actively surfacing scattered decision states.

OPSD-V: On-Policy Self-Distillation for Post-Training Few-Step Autoregressive Video Generators

AI·OPSD-V self-distillation method reduces error accumulation in few-step autoregressive video generators while preserving fast inference.

ARDY: Autoregressive Diffusion with Hybrid Representation for Interactive Human Motion Generation

AI·ARDY hybrid autoregressive diffusion model generates interactive real-time 3D human motions for animation and humanoid robotics.

Dual Latent Memory in Vision-Language-Action Models for Robotic Manipulation

AI·Dual Latent Memory in Vision-Language-Action models for robotic manipulation interleaves historical experience fluidly in the native latent embedding space, overcoming Markovian limitations in long-horizon tasks.

Single-Rollout Asynchronous Optimization for Agentic Reinforcement Learning

AI·Single-Rollout Asynchronous Optimization is an agentic RL method that replaces group-wise sampling with single-rollout updates and adds token-level clipping to improve stability and performance on coding and reasoning benchmarks.

RoboTALES: Learning Reasoning-Guided Robot Policies via Task-Aligned Simulated Futures

AI·RoboTALES trains robot policies by learning task-aligned simulated futures with a hierarchical LLM-based planner and VLM-based critic, producing coherent actions that outperform prior methods on long-horizon manipulation benchmarks from RoboCasa and LIBERO10.

Imagined Rollouts are Kinematic, Not Dynamic: A Diagnosis of Long-Horizon World-Model Failure

Long-horizon failure in world models is conventionally attributed to compounding error, a generic framing that does not distinguish what kind of error compounds. We propose a kinematic-vs-dynamic reframing: world models tend to imagine kinematically rather than dynamically. We operationalize this as the imagined Kinematic-Consistency Error, a per-step diagnostic that measures how far a rollout departs from a closed-form kinematic null, paired with a perturbation protocol that tests whether iKCE…