Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

Individual utilities of life satisfaction reveal inequality aversion unrelated to political alignment

DischargeSim: A Simulation Benchmark for Educational Doctor-Patient Communication at Discharge

Moment- and Power-Spectrum-Based Gaussianity Regularization for Text-to-Image Models

Computational Concept of the Psyche (in Russian)

MachineLearningLM: Scaling Many-shot In-context Learning via Continued Pretraining

The Efficiency Frontier: Classical Shadows versus Quantum Footage

BranchGRPO: Stable and Efficient GRPO with Structured Branching in Diffusion Models

Uncertainty Quantification in Probabilistic Machine Learning Models: Theory, Methods, and Insights

CURE: Controlled Unlearning for Robust Embeddings - Mitigating Conceptual Shortcuts in Pre-Trained Language Models

Revealing Hidden Precursors to Earthquakes via a Stress-Sensitive Transformation of Seismic Noise

ASE: A Repository-Level Benchmark for Evaluating Security in AI-Generated Code

Subjective Behaviors and Preferences in LLM: Language of Browsing

Beyond Ten Turns: Unlocking Long-Horizon Agentic Search with Large-Scale Asynchronous RL

SGDFuse: SAM-Guided Diffusion for High-Fidelity Infrared and Visible Image Fusion

Self-Questioning Language Models

MetaExplainer: A Framework to Generate Multi-Type User-Centered Explanations for AI Systems

How Should We Meta-Learn Reinforcement Learning Algorithms?

Comprehensive Evaluation of Prototype Neural Networks

HIRAG: Hierarchical-Thought Instruction-Tuning Retrieval-Augmented Generation

CyberRAG: An Agentic RAG cyber attack classification and reporting tool

Multi-Timescale Hierarchical Reinforcement Learning for Unified Behavior and Control of Autonomous Driving

A Nonlinear Low-rank Representation Model with Convolutional Neural Network for Imputing Water Quality Data

VIDEE: Visual and Interactive Decomposition, Execution, and Evaluation of Text Analytics with Intelligent Agents

Discrete Diffusion in Large Language and Multimodal Models: A Survey

From Static to Adaptive Defense: Federated Multi-Agent Deep Reinforcement Learning-Driven Moving Target Defense Against DoS Attacks in UAV Swarm Networks

How Far Are We from Optimal Reasoning Efficiency?

Whose Name Comes Up? Auditing LLM-Based Scholar Recommendations

Stopping Criteria for Value Iteration on Concurrent Stochastic Reachability and Safety Games

Your Language Model Can Secretly Write Like Humans: Contrastive Paraphrase Attacks on LLM-Generated Text Detectors

Prior Prompt Engineering for Reinforcement Fine-Tuning

Reasoning Large Language Model Errors Arise from Hallucinating Critical Problem Features

CoT-RAG: Integrating Chain of Thought and Retrieval-Augmented Generation to Enhance Reasoning in Large Language Models

TransitReID: Transit OD Data Collection with Occlusion-Resistant Dynamic Passenger Re-Identification

TerraMind: Large-Scale Generative Multimodality for Earth Observation

Recursive Training Loops in LLMs: How training data properties modulate distribution shift in generated data?

Scaling Video-Language Models to 10K Frames via Hierarchical Differential Distillation

A decision-theoretic approach to dealing with uncertainty in quantum mechanics

VIPER: Visual Perception and Explainable Reasoning for Sequential Decision-Making

LED: LLM Enhanced Open-Vocabulary Object Detection without Human Curated Data Generation

Reangle-A-Video: 4D Video Generation as Video-to-Video Translation

To See a World in a Spark of Neuron: Disentangling Multi-task Interference for Training-free Model Merging

UAR-NVC: A Unified AutoRegressive Framework for Memory-Efficient Neural Video Compression

MPO: Boosting LLM Agents with Meta Plan Optimization

Pay Attention to Real World Perturbations! Natural Robustness Evaluation in Machine Reading Comprehension

A general language model for peptide identification

Beyond Seen Data: Improving KBQA Generalization Through Schema-Guided Logical Form Generation

CoAT: Chain-of-Associated-Thoughts Framework for Enhancing Large Language Models Reasoning

Mind the Value-Action Gap: Do LLMs Act in Alignment with Their Values?

Traffic-Rule-Compliant Trajectory Repair via Satisfiability Modulo Theories and Reachability Analysis

QR-VC: Leveraging Quantization Residuals for Linear Disentanglement in Zero-Shot Voice Conversion

Generative AI for Data Augmentation in Wireless Networks: Analysis, Applications, and Case Study

Neural-Enhanced Dynamic Range Compression Inversion: A Hybrid Approach for Restoring Audio Dynamics

The Quest for the Right Mediator: Surveying Mechanistic Interpretability Through the Lens of Causal Mediation Analysis

PriorCLIP: Visual Prior Guided Vision-Language Model for Remote Sensing Image-Text Retrieval

A Transformer approach for Electricity Price Forecasting

FedComLoc: Communication-Efficient Distributed Training of Sparse and Quantized Models

PQMass: Probabilistic Assessment of the Quality of Generative Models using Probability Mass Estimation

HiPhO: How Far Are (M)LLMs from Humans in the Latest High School Physics Olympiad Benchmark?

Towards explainable decision support using hybrid neural models for logistic terminal automation

BlendedNet: A Blended Wing Body Aircraft Dataset and Surrogate Model for Aerodynamic Predictions

That's So FETCH: Fashioning Ensemble Techniques for LLM Classification in Civil Legal Intake and Referral

Murphys Laws of AI Alignment: Why the Gap Always Wins

Adaptive Monitoring and Real-World Evaluation of Agentic AI Systems

Bridging the Gap in Ophthalmic AI: MM-Retinal-Reason Dataset and OphthaReason Model toward Dynamic Multimodal Reasoning

Understanding visual attention beehind bee-inspired UAV navigation

Working with AI: Measuring the Applicability of Generative AI to Occupations

Scaling LLM Planning: NL2FLOW for Parametric Problem Generation and Rigorous Evaluation

Context-Driven Knowledge Graph Completion with Semantic-Aware Relational Message Passing

Meta-Semantics Augmented Few-Shot Relational Learning

Perovskite-LLM: Knowledge-Enhanced Large Language Models for Perovskite Solar Cell Research

Associative Knowledge Graphs for Efficient Sequence Storage and Retrieval

Depth-Bounded Epistemic Planning

A Survey of Reinforcement Learning for Large Reasoning Models

Large Language Model Hacking: Quantifying the Hidden Risks of Using LLMs for Text Annotation

QCardEst/QCardCorr: Quantum Cardinality Estimation and Correction

Merge-of-Thought Distillation

MoVoC: Morphology-Aware Subword Construction for Geez Script Languages

Scaling Truth: The Confidence Paradox in AI Fact-Checking

PianoVAM: A Multimodal Piano Performance Dataset

An End-to-End Deep Learning Framework for Arsenicosis Diagnosis Using Mobile-Captured Skin Images

Using AI to Optimize Patient Transfer and Resource Utilization During Mass-Casualty Incidents: A Simulation Platform

AgentGym-RL: Training LLM Agents for Long-Horizon Decision Making through Multi-Turn Reinforcement Learning

Learning Turbulent Flows with Generative Models: Super-resolution, Forecasting, and Sparse Flow Reconstruction

FinZero: Launching Multi-modal Financial Time Series Forecast with Large Reasoning Model

DEQuify your force field: More efficient simulations using deep equilibrium models

X-Teaming Evolutionary M2S: Automated Discovery of Multi-turn to Single-turn Jailbreak Templates

Explainability of CNN Based Classification Models for Acoustic Signal

TANGO: Traversability-Aware Navigation with Local Metric Control for Topological Goals

A layered architecture for log analysis in complex IT systems

Reshaping the Forward-Forward Algorithm with a Similarity-Based Objective

Skeleton-based sign language recognition using a dual-stream spatio-temporal dynamic graph convolutional network

Robust Belief-State Policy Learning for Quantum Network Routing Under Decoherence and Time-Varying Conditions

Architecting Resilient LLM Agents: A Guide to Secure Plan-then-Execute Implementations

RoentMod: A Synthetic Chest X-Ray Modification Model to Identify and Correct Image Interpretation Model Shortcuts

UOPSL: Unpaired OCT Predilection Sites Learning for Fundus Image Diagnosis Augmentation

OTESGN:Optimal Transport Enhanced Syntactic-Semantic Graph Networks for Aspect-Based Sentiment Analysis

Classification of 24-hour movement behaviors from wrist-worn accelerometer data: from handcrafted features to deep learning techniques

Memorization in Large Language Models in Medicine: Prevalence, Characteristics, and Implications

Interpretability as Alignment: Making Internal Understanding a Design Principle

MESH -- Understanding Videos Like Humans: Measuring Hallucinations in Large Video Models

Murphys Laws of AI Alignment: Why the Gap Always Wins

Created by

Haebom

Author

Madhava Gaikwad

Outline

This paper demonstrates a formal impossibility result for Reinforcement Learning from Human Feedback (RLHF). In a poorly specified environment with a limited query budget, RLHF-style learners suffer from an unbridgeable performance gap Omega(γ) without access to a correction oracle. Information-theoretic proofs provide strict lower bounds and demonstrate that a minimal number of correction oracles is sufficient to close the gap. A small empirical example and a list of alignment rules (Murphy's Law) demonstrate that many observed alignment failures are consistent with this structural mechanism. These findings establish Murphy's gap as a diagnostic limitation of RLHF and serve as a guide for future research on correction and causal preference identification.

Takeaways, Limitations

•

Takeaways: We identify "Murphy's Gap," a structural limitation of RLHF, and suggest the importance of a correction oracle to address it. We provide information-theoretic evidence proving the performance limitations of RLHF in poorly specified environments, suggesting future directions for RLHF research. We also provide a new explanation for the observed alignment failures.

•

Limitations: Only small-scale empirical examples are presented, and further research is needed to determine the applicability and generalizability to large-scale systems. There is a lack of specific discussion on the implementation and practical application of correction oracles. The definition and scope of "Murphy's Law" are unclear, requiring further explanation.

Made with Slashpage