AI研究论文每日精选

Reinforcement Pre-Training

Lingshu: A Generalist Foundation Model for Unified Multimodal Medical Understanding and Reasoning

Saffron-1: Towards an Inference Scaling Paradigm for LLM Safety Assurance

MiniCPM4: Ultra-Efficient LLMs on End Devices

OneIG-Bench: Omni-dimensional Nuanced Evaluation for Image Generation

SpatialLM: Training Large Language Models for Structured Indoor Modeling

Image Reconstruction as a Tool for Feature Analysis

Astra: Toward General-Purpose Mobile Robots via Hierarchical Multimodal Learning

Pre-trained Large Language Models Learn Hidden Markov Models In-context

Rethinking Cross-Modal Interaction in Multimodal Diffusion Transformers

GTR-CoT: Graph Traversal as Visual Chain of Thought for Molecular Structure Recognition

BitVLA: 1-bit Vision-Language-Action Models for Robotics Manipulation

Debatable Intelligence: Benchmarking LLM Judges via Debate Speech Evaluation

Through the Valley: Path to Effective Long CoT Training for Small Language Models

The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity

Bootstrapping World Models from Dynamics Models in Multimodal Foundation Models

Vision Transformers Don't Need Trained Registers

CCI4.0: A Bilingual Pretraining Dataset for Enhancing Reasoning in Large Language Models

ConfQA: Answer Only If You Are Confident

Model Immunization from a Condition Number Perspective

Play to Generalize: Learning to Reason Through Game Play

GUI-Reflection: Empowering Multimodal GUI Models with Self-Reflection Behavior

Well Begun is Half Done: Low-resource Preference Alignment by Weak-to-Strong Decoding

Dreamland: Controllable World Creation with Simulator and Generative Models

SynthesizeMe! Inducing Persona-Guided Prompts for Personalized Reward Models in LLMs

ExpertLongBench: Benchmarking Language Models on Expert-Level Long-Form Generation Tasks with Structured Checklists

What Is Seen Cannot Be Unseen: The Disruptive Effect of Knowledge Conflict on Large Language Models

Cartridges: Lightweight and general-purpose long context representations via self-study

Agents of Change: Self-Evolving LLM Agents for Strategic Planning

τ^2-Bench: Evaluating Conversational Agents in a Dual-Control Environment

SAFEFLOW: A Principled Protocol for Trustworthy and Transactional Autonomous Agent Systems

Learning What Reinforcement Learning Can't: Interleaved Online Fine-Tuning for Hardest Questions

Self-Adapting Improvement Loops for Robotic Learning

NetPress: Dynamically Generated LLM Benchmarks for Network Applications

PolyVivid: Vivid Multi-Subject Video Generation with Cross-Modal Interaction and Enhancement

Overclocking LLM Reasoning: Monitoring and Controlling Thinking Path Lengths in LLMs

GeometryZero: Improving Geometry Solving for LLM with Group Contrastive Policy Optimization

MegaHan97K: A Large-Scale Dataset for Mega-Category Chinese Character Recognition with over 97K Categories

Robust Preference Optimization via Dynamic Target Margins

Dynamic View Synthesis as an Inverse Problem

CyberV: Cybernetics for Test-time Scaling in Video Understanding

Improving large language models with concept-aware fine-tuning

Evaluating LLMs Robustness in Less Resourced Languages with Proxy Models

Training-Free Tokenizer Transplantation via Orthogonal Matching Pursuit

Hidden in Plain Sight: Probing Implicit Reasoning in Multimodal Language Models

EVOREFUSE: Evolutionary Prompt Optimization for Evaluation and Mitigation of LLM Over-Refusal to Pseudo-Malicious Instructions

Meta-Adaptive Prompt Distillation for Few-Shot Visual Question Answering

Proactive Assistant Dialogue Generation from Streaming Egocentric Videos