AI summaries of arXiv papers
Paste any arXiv URL or paper ID. Or just swap arxiv.org → arxivtldr.org in your address bar.
Add to Chrome — Free ExtensionToday's Papers
Rubric-CEPR: Self-Evolving Image Editing via Reward-Verified Self-Distillation
Rubric-CEPR is a self-evolving framework that improves image editors using their own generations, verified by an internal rubric-augmented reward.
Dex-One2Many: Learning Dexterous Manipulation from a Single Human Demonstration
Dex-One2Many learns generalizable dexterous manipulation from a single human video by abstracting it into scene graphs to guide reinforcement learning.
DreamTrue: Action-Faithful Robot World Model with Counterfactual Post-Training
DreamTrue is a robot world model that uses counterfactual post-training and geometric calibration to improve action-faithful and physically plausible video prediction.
A Balanced Data Diet: Addressing the Exploration Bottleneck in Mega-Scale RL for Robot Control
This paper introduces Success Guided Sampling (SGS) to optimize large-scale RL exploration for complex robot control tasks.
CSF: Contextual Safety Filtering for Motion Generators
CSF is a training-free filter for motion generators that uses natural-language safety rules and reference trajectories to prevent unsafe actions.
On the estimation and validity of AI time horizons---a statistical look at the METR plot
This paper statistically re-evaluates AI time horizons, finding a non-linear relationship between human time and AI difficulty, improving estimation and validity.
Disk Structure May Determine AGN Variability and Explain the Accretion Disk Size Problem: No Broad Line Region Required
Disk structure, specifically thickness, significantly impacts AGN variability and can explain observed discrepancies in accretion disk size and reverberation lags.
From Reactive Containment to Proactive Assurance: Lessons from OpenAI, Anthropic, and Google Agent Security Incidents
This paper analyzes agent security incidents at OpenAI, Anthropic, and Google, proposing a Proactive Agent Security Assurance Cycle (PASAC).
What 30,000 Hours of Ego-centric Video Does Not Teach
Scaling ego-centric video improves agent modeling but struggles with object fidelity, suggesting architectural changes are needed beyond just more data.
OuroWorld: Bringing Any 3D World Alive as Diverse, Endlessly Looping 3D Cinemagraphs
OuroWorld transforms static 3D Gaussian Splatting scenes into dynamic, endlessly looping 3D cinemagraphs with diverse motion from any viewpoint.
WorldGuide: Goal-Directed Video World Model for Procedural Task Execution
WorldGuide is a novel video world model that performs closed-loop, goal-directed procedural task execution by coupling planning with learned execution.
OmniCapBench: A Deep-Structured Evaluation Framework for Fine-Grained Audio-Visual Captioning
OmniCapBench is a new deep-structured benchmark for fine-grained audio-visual captioning, revealing MLLM limitations in temporal grounding and cross-modal alignment.
Browse by Category
Browse by author →Browse by Topic
All topics →📬 Weekly AI Paper Digest
Get the top 10 AI/ML arXiv papers from the week — summarized, scored, and delivered to your inbox every Monday.
The URL swap trick
Reading a paper on arxiv.org? Just change the domain to arxivtldr.org in your address bar:
Paste a link
Enter any arXiv URL or paper ID
AI summarizes
Get a TLDR, key bullets, and why it matters
Read & share
Get key insights in seconds, share with colleagues