This repository curates a comprehensive collection of resources on Large Language Model (LLM) safety, targeting researchers and practitioners. It provides a structured overview of papers, tutorials, articles, and datasets covering security, privacy, truthfulness, jailbreaking, defenses, and benchmarks, aiming to keep users updated on the latest advancements and challenges in LLM safety.

How It Works

The repository organizes information into distinct categories such as Security & Discussion, Privacy, Truthfulness & Misinformation, Jailbreak & Attacks, Defenses & Mitigation, and Datasets & Benchmarks. Within each category, resources are further categorized into papers and tutorials/articles/presentations, with links and brief descriptions provided. The project is actively updated with new findings and aims to be a central hub for LLM safety research.

Quick Start & Requirements

• Access resources directly via the README or explore subtopic folders for deeper dives.
• No specific installation or computational requirements are mentioned, as it's a curated list of external resources.
• Links to official quick-start guides or demos are not provided, but the README itself serves as a guide.

Highlighted Details

• Covers a broad spectrum of LLM safety topics, including emerging areas like jailbreaking and adversarial attacks.
• Actively updated with recent publications, including NAACL 2024 papers.
• Includes links to relevant talks, conferences, and news articles in addition to academic papers.
• Provides a curated list of select information sorted by date for quick access.

Maintenance & Community

• Contributions are welcomed via GitHub issues for individual papers or pull requests for compiled conference papers.
• A WeChat group is available for LLM Safety discussions.
• Contact email provided for inquiries and promotional opportunities.

Licensing & Compatibility

• The repository itself is not software and thus not subject to software licensing. The linked resources will have their own respective licenses.
• Compatibility for commercial use or closed-source linking depends entirely on the licenses of the individual resources linked within the repository.

Limitations & Caveats

The repository is a curated list and does not host any code or models itself. The quality and relevance of the information are dependent on the contributors and the ongoing maintenance of the links.