Trending
What is gaining attention fastest, ranked by growth in upvotes as reported by the source, plus brand-new entries that arrived with traction. Looking for everything that just appeared, in order? See New.
- Any Time, Any Space: A World-State-Centric Framework for Agentic Video Generation New+330 upvotes
Long-form agentic video generation must turn a screenplay into shots that remain spatially and temporally consistent across nonconsecutive requests. Passing pro
- Doc2Harness: Benchmarking LLMs on Long-Horizon Harness Engineering of Coding Agents New+96 upvotes
Agents driven by large language models are increasingly deployed on long-horizon tasks, where they use tools and modify real workspaces. How well they perform d
- OmniCompass: A Holistic Benchmark for Omni-Modal Video Grounding and Understanding New+90 upvotes
With the rapid development of multimodal large language models (MLLMs), evaluating their video understanding capabilities has become increasingly important. Exi
- More Methods, Better Agents? A Reality Check on Self-Distillation in Agentic Reinforcement Learning New+76 upvotes
Self-distillation in agentic reinforcement learning has produced a growing number of methods. However, existing evaluations often reuse published baseline score
- Recovering Reliable Supervision via Critique-Guided On-Policy Distillation New+76 upvotes
On-policy distillation (OPD) trains a student model on its own rollouts with token-level supervision from a teacher, assuming that the teacher can reach a corre
- Cortex: Consolidating Context into Parametric Memory New+65 upvotes
Context can improve large language model performance, but these gains may be lost after context removal. Inspired by biological systems consolidation, we formul
- LatentLoop: Scaling Looped MoE Models with Asymmetric Compute Allocation New+65 upvotes
Looped Mixture-of-Experts (MoE) models combine parameter reuse with sparse computation to improve performance under matched budgets. Prior work demonstrates tha
- RuleTrek: Benchmarking Agentic Rule Reasoning in Complex Scenarios New+63 upvotes
Large language models (LLMs) increasingly tackle complex real-world tasks in specialized domains. However, identifying and correctly applying domain-specific ru
- Self-Evolving Agent Harness New+63 upvotes
Harness self-evolution revises an agent's software scaffolding, including prompts, memory, workflows, and executable components, while keeping model weights fix
- Speculative Forcing for Efficient Autoregressive Video Generation New+56 upvotes
Few-step autoregressive video generators enable streaming generation, yet they run every token through the final denoising step, regardless of how much refineme
- ActionOPD: Distilling Decisions, Not Tokens New+54 upvotes
Distilling capable language-model agents into smaller models requires aligning the student's behavior with the teacher's. This raises a fundamental question: at
- Waypoint-OPD: Transferring RL-Induced Policy Shifts via Gated On-Policy Distillation New+49 upvotes
Policy-shift distillation reuses the difference between a teacher's log probabilities before and after reinforcement learning (RL) as supervision on student-gen
- Seeing, Maintaining, and Learning: The Evolution of Multimodal Agent Memory New+46 upvotes
Multimodal models and agents have emerged to enable AI systems to learn from and interact with the real world more like humans by integrating information across
- Many Right Answers, One Training Example: Removing Responder Noise from GRPO for LLM Recommenders New+39 upvotes
A recommendation request on Reddit draws several replies, each naming different titles. When an LLM recommender is trained on such threads with reinforcement le
- LensErase: A Foundation Model for Erase Anything New+37 upvotes
We introduce LensErase, a general-purpose mask-based Erase Anything foundation model designed to cover a broad spectrum of real-world erase scenarios. LensErase
- SCOPE: Semantic Bounded Credit Optimization for Prompt Engineering in Text-to-image Generation New+36 upvotes
Reinforcement learning (RL) enables language models to improve text-to-image (T2I) generation by rewriting prompts for a frozen image generator. Existing method
- Apertus 1.5: From Open Text Models to Open Multimodal Reasoners New+35 upvotes
Language models are becoming infrastructure, and infrastructure built in the open, by public institutions, on documented data, is the alternative to dependence
- KV4-RL: Unlocking Faster Asynchronous Rollouts for RLVR with INT4 KV Cache Quantization New+32 upvotes
Asynchronous rollout improves the efficiency of reinforcement learning with verifiable rewards (RLVR) by reducing under-utilization during long-tail generation.
- Learning Where to Split: Dynamic Band Splitting for Audio Source Separation New+32 upvotes
Audio source separation models must preserve fine spectral structure while keeping the cost of the separator tractable. Band-split architectures reduce this cos
- CLASHarness: Benchmarking Multi-Agent Conflict in Computer Use via a Controllable and Verifiable Generation Pipeline New+30 upvotes
Vision-language models (VLMs) are increasingly deployed in multi-agent systems to automate complex computer-use workflows, yet existing benchmarks mostly assume
- Generative World Simulation with DAO and JING New+9 upvotes
We present a Generative World Simulation framework that couples persistent shared world state with continuous first-person audiovisual generation. The framework
- Share the Symbol, Not the Vector: Training-Free Fact Communication Across Language Models New+9 upvotes
Multi-agent systems allow language models to solve tasks collaboratively by sharing complementary information. These systems typically communicate through text,
- ProbabilityBench: Model Evaluation and Calibration with Exact Ground-Truth Uncertainty New+9 upvotes
Evaluating model uncertainty on real-world data is challenging because each instance provides only a single observed label, leaving true conditional class proba
- What Makes a Social Agent's Interactions Effective? Learning from Verifiable Outcomes in Paired Worlds New+9 upvotes
Effective social interaction requires more than fluent responses: agents must reach plans that fit the participants' needs and circumstances. Existing evaluatio
- Athanor: An Agent-Native System for Autonomous Post-Training Research New+8 upvotes
Recursive self-improvement, in which AI systems build stronger AI systems, is one promising route to superintelligence. We study a weaker form that can be measu
- Glance Forcing: Making Bidirectional Diffusion Autoregressive at a Glance New+8 upvotes
Autoregressive generation underlies real-time, causal video synthesis and is the backbone of video world models. Today's dominant approach distills a bidirectio
- VITAL: From Visual Motion to Precise Action New+8 upvotes
World-action models aim to turn visual anticipation into driving decisions, yet typically train action experts on trajectory representations defined independent
- From Failures to Learnable Worlds: Policy-Conditioned Environment Generation for Navigation Agents New+8 upvotes
Simulated environments enable scalable training data generation for embodied agents, but more data does not necessarily provide what a particular agent needs to
- WorldMemBench: A Comprehensive Benchmark for Memory of Video World Models New+8 upvotes
As video models enable longer and more realistic interactive experiences, maintaining a world consistent with its history becomes essential. This requires both
- TextVerifier: Seeing Beyond Recognition to Verify Chinese Text Rendering New+8 upvotes
Text-to-image models increasingly generate readable Chinese text that still contains local rendering defects, such as a character with an extra or missing strok
Just indexed
- Connecting…
Counts are the numbers reported by Hugging Face, GitHub and arXiv-linked communities, recorded daily by Silytics. Downloads on Hugging Face are 30-day figures, so a model that is already popular can still rank. Papers are ranked by Hugging Face daily-paper upvotes.