Why_this_webapp Bibliography

A Survey of Safety and Trustworthiness of Large Language Models through the Lens of Verification and Validation / 2305.11391 / ISBN:https://doi.org/10.48550/arXiv.2305.11391 / Published by ArXiv / on (web) Publishing site

2 Large Language Models

Specific versus General Principles for Constitutional AI / 2310.13798 / ISBN:https://doi.org/10.48550/arXiv.2310.13798 / Published by ArXiv / on (web) Publishing site

Contents
1 Introduction
2 AI feedback on specific problematic AI traits
3 Generalization from a Simple Good for Humanity Principle
4 Reinforcement Learning with Good-for-Humanity Preference Models
7 Contribution Statement
A Model Glossary

AI Alignment and Social Choice: Fundamental Limitations and Policy Implications / 2310.16048 / ISBN:https://doi.org/10.48550/arXiv.2310.16048 / Published by ArXiv / on (web) Publishing site

1 Introduction
3 Arrow-Sen Impossibility Theorems for RLHF

Human participants in AI research: Ethics and transparency in practice / 2311.01254 / ISBN:https://doi.org/10.48550/arXiv.2311.01254 / Published by ArXiv / on (web) Publishing site

IV. Principles in Practice: Guidelines for AI Research with Human Participants

How Trustworthy are Open-Source LLMs? An Assessment under Malicious Demonstrations Shows their Vulnerabilities / 2311.09447 / ISBN:https://doi.org/10.48550/arXiv.2311.09447 / Published by ArXiv / on (web) Publishing site

4 Experiments

Case Repositories: Towards Case-Based Reasoning for AI Alignment / 2311.10934 / ISBN:https://doi.org/10.48550/arXiv.2311.10934 / Published by ArXiv / on (web) Publishing site

3 Related Work and Discussion

Control Risk for Potential Misuse of Artificial Intelligence in Science / 2312.06632 / ISBN:https://doi.org/10.48550/arXiv.2312.06632 / Published by ArXiv / on (web) Publishing site

3 Control the Risks of AI Models in Science
Appendix D Details of Benchmark Results

Inadequacies of Large Language Model Benchmarks in the Era of Generative Artificial Intelligence / 2402.09880 / ISBN:https://doi.org/10.48550/arXiv.2402.09880 / Published by ArXiv / on (web) Publishing site

V. Processual Elements

AI Alignment: A Comprehensive Survey / 2310.19852 / ISBN:https://doi.org/10.48550/arXiv.2310.19852 / Published by ArXiv / on (web) Publishing site

2 Learning from Feedback
4 Assurance
5 Governance

Large Language Model Supply Chain: A Research Agenda / 2404.12736 / ISBN:https://doi.org/10.48550/arXiv.2404.12736 / Published by ArXiv / on (web) Publishing site

4 LLM Lifecycle

The AI Alignment Paradox / 2405.20806 / ISBN:https://doi.org/10.48550/arXiv.2405.20806 / Published by ArXiv / on (web) Publishing site

Paper

How Ethical Should AI Be? How AI Alignment Shapes the Risk Preferences of LLMs / 2406.01168 / ISBN:https://doi.org/10.48550/arXiv.2406.01168 / Published by ArXiv / on (web) Publishing site

Abstract
Introduction
I. Description of Method/Empirical Design
III. Impact of Alignment on LLMs’ Risk Preferences
V. Conclusions
Figures and tables

AI Alignment through Reinforcement Learning from Human Feedback? Contradictions and Limitations / 2406.18346 / ISBN:https://doi.org/10.48550/arXiv.2406.18346 / Published by ArXiv / on (web) Publishing site

Abstract
3 Limitations of RLxF
4 The Internal Tensions and Ethical Issues in RLxF

Thorns and Algorithms: Navigating Generative AI Challenges Inspired by Giraffes and Acacias / 2407.11360 / ISBN:https://doi.org/10.48550/arXiv.2407.11360 / Published by ArXiv / on (web) Publishing site

Abstract
7 Recommendations: Fixing Gen AI’s Value Alignment
8 Conclusion

DailyDilemmas: Revealing Value Preferences of LLMs with Quandaries of Daily Life / 2410.02683 / ISBN:https://doi.org/10.48550/arXiv.2410.02683 / Published by ArXiv / on (web) Publishing site

1 Introduction

From human-centered to social-centered artificial intelligence: Assessing ChatGPT's impact through disruptive events / 2306.00227 / ISBN:https://doi.org/10.48550/arXiv.2306.00227 / Published by ArXiv / on (web) Publishing site

The multiple levels of AI impact

Navigating the Cultural Kaleidoscope: A Hitchhiker's Guide to Sensitivity in Large Language Models / 2410.12880 / ISBN:https://doi.org/10.48550/arXiv.2410.12880 / Published by ArXiv / on (web) Publishing site

7 Cultural safeguarding

Hybrid Approaches for Moral Value Alignment in AI Agents: a Manifesto / 2312.01818 / ISBN:https://doi.org/10.48550/arXiv.2312.01818 / Published by ArXiv / on (web) Publishing site

3. Designing AI Agents based on Moral Principles

Safety at Scale: A Comprehensive Survey of Large Model Safety / 2502.05206 / ISBN:https://doi.org/10.48550/arXiv.2502.05206 / Published by ArXiv / on (web) Publishing site

3 Large Language Model Safety

Position: We Need An Adaptive Interpretation of Helpful, Honest, and Harmless Principles / 2502.06059 / ISBN:https://doi.org/10.48550/arXiv.2502.06059 / Published by ArXiv / on (web) Publishing site

Abstract
1 Introduction
2 HHH Principle
3 Ambiguity and Conflicts in HHH
4 Priority Order
5 Trade-off or Synergy? Relationship Between Different Dimensions
6 Reference Framework
7 Open Challenge
Appendices

On the Trustworthiness of Generative Foundation Models: Guideline, Assessment, and Perspective / 2502.14296 / ISBN:https://doi.org/10.48550/arXiv.2502.14296 / Published by ArXiv / on (web) Publishing site

3 Guidelines of Trustworthy Generative Foundation Models
9 Trustworthiness in Downstream Applications

DarkBench: Benchmarking Dark Patterns in Large Language Models / 2503.10728 / ISBN:https://doi.org/10.48550/arXiv.2503.10728 / Published by ArXiv / on (web) Publishing site

Referemces

Values in the Wild: Discovering and Analyzing Values in Real-World Language Model Interactions / 2504.15236 / ISBN:https://doi.org/10.48550/arXiv.2504.15236 / Published by ArXiv / on (web) Publishing site

1 Introduction

Analysing Safety Risks in LLMs Fine-Tuned with Pseudo-Malicious Cyber Security Data / 2505.09974 / ISBN:https://doi.org/10.48550/arXiv.2505.09974 / Published by ArXiv / on (web) Publishing site

2 Related Work

Just as Humans Need Vaccines, So Do Models: Model Immunization to Combat Falsehoods / 2505.17870 / ISBN:https://doi.org/10.48550/arXiv.2505.17870 / Published by ArXiv / on (web) Publishing site

Appendix

Wide Reflective Equilibrium in LLM Alignment: Bridging Moral Epistemology and AI Safety / 2506.00415 / ISBN:https://doi.org/10.48550/arXiv.2506.00415 / Published by ArXiv / on (web) Publishing site

1. Introduction: The Convergence of Moral Epistemology and AI Safety
4. The Landscape of LLM Alignment: Methods and Challenges
5. Wide Reflective Equilibrium as the Descriptive Key to LLM Alignment

DeepSeek in Healthcare: A Survey of Capabilities, Risks, and Clinical Applications of Open-Source Large Language Models / 2506.01257 / ISBN:https://doi.org/10.48550/arXiv.2506.01257 / Published by ArXiv / on (web) Publishing site

Appendix