arXiv
Verdict: NotableRecurrent Looped Transformer
State tracking requires an update at every input, but the depth a Transformer applies to each token is fixed regardless of sequence length.
- HF ▲ 26
- Stars ★ 918
Artificial Intelligence
Models, methods, and the machinery of intelligence.
Edition No. 11
Updated UTC
Top stories
1. Kandinsky 6.0 Video: Foundation Models for Synchronized Video and Audio Generation
We present Kandinsky 6.0 Video, a family of foundation diffusion models for synchronized text-to-audio-video generation, comprising Kandinsky 6.0 Video Lite (3B parameters) and Kandinsky 6.0 Video Pro (29B parameters).
██████░░░░5.52. Long-WAM: Scaling the Context of World-Action Models
Real-time robot control demands enough visual history to infer motion and task progress, but processing that history can delay action.
██████░░░░5.53. SuperNav: An Agentic Navigation System for Any Task in Any Scene
General-purpose service robots need navigation systems that can handle diverse human requests in unfamiliar environments, combining task generality with scene generality.
█████░░░░░5.4
More top stories
- arXivby Liu, Liu, Zhang
nanoMuse: An Open-Source Personal Agent for Every Device You Own
Assistants from 2011 answered and waited, and agents from 2023 did a task and stopped.
No.04Verdict: RoutineScore█████░░░░░5.4 - Hugging Faceby Su, Zhuang, Li +10
SGF+: Decoupling Gradient Flows for Autoregressive Video Generation
Autoregressive video generation requires denoising the current frames while writing their key-value representations as context for future predictions.
No.05Verdict: RoutineScore█████░░░░░5.3 - Hugging Faceby Yang, Li, Hu +30
RobotWorld: Benchmarking Multimodal Agents for Robot Use Across Diverse Tasks and Embodiments
General-purpose agents increasingly write code, use tools, and complete complex digital tasks, raising the question of how far these capabilities carry into the physical world.
No.06Verdict: NotableScore█████░░░░░5.3 - arXivby Wang, Yu, Zhuge +9
TRACE: Rollout-Guided Quantization-Aware Training for FP4 Reinforcement Learning of MoE Language Models
Reinforcement learning (RL) for post-training large language models (LLMs) incurs substantial computation and memory overhead during rollout generation, which motivates low-precision ....
No.07Verdict: RoutineScore█████░░░░░5.3 - arXivby Zhang, Zhang, Hu +5
LoGRA: Scaling LLM Reinforcement Learning with Low-Rank Gradient Sketches
Reinforcement learning (RL) has greatly advanced the capabilities of large language models (LLMs), but its memory demands remain a barrier to broader adoption.
No.08Verdict: RoutineScore█████░░░░░5.3
Briefs
- arXiv
Recursive Game Creator: An Agentic Product-Level Experience-Oriented Game Harness
█████░░░░░5.3 - Hugging Face
GRACE: Generation-aware latent compression for efficient video generation
█████░░░░░5.3 - arXiv
Rethinking Cross-Tokenizer On-Policy Distillation: From Alignment Coverage to Supervision Reliability
█████░░░░░5.3 - arXiv
HuatuoGPT-3: RL-Only Domain Adaptation from Base Models
█████░░░░░5.2 - Hugging Face
Tetris3D: 3D Scene Generation With Objects That Fit Together
█████░░░░░5.2 - arXiv
Memadapter: Counterfactual Adaptation Against Memory-induced Sycophancy
█████░░░░░5.1 - arXiv
From Evidence to Action: How Tool-Using Agents Fail
█████░░░░░5.1 - arXiv
NeMo-DCR: Bit-Exact Delta-Compressed Refit for Scalable Agentic RL at Trillion-Parameter Scale
█████░░░░░5.1 - arXiv
Self-Retrospection Distillation: Turning Post-hoc Experiences into Prior Foresight
█████░░░░░5.1 - arXiv
DecepEval: A Benchmark for Evaluating Deception in LLM Agents
█████░░░░░5.1