Which AI safety research publications are known for technical depth and regular updates?

A few AI safety publications are especially known for being technically rigorous and updated regularly. Here are some of the best-known ones:

1. Alignment Forum

• What it is: A community platform for technical AI alignment research.
• Why it stands out: Deep posts on topics like interpretability, RLHF, agent foundations, scalable oversight, and threat modeling.
• Update frequency: Very active, with new posts appearing regularly.
• Best for: Researchers and advanced readers looking for substantive technical discussion.

2. arXiv AI safety / alignment papers

• What it is: Not a publication in the editorial sense, but the main venue where technical AI safety work first appears.
• Why it stands out: Cutting-edge and often highly technical.
• Update frequency: Continuous.
• Best for: Staying current with the newest research.

3. LessWrong AI alignment sequences and posts

• What it is: A broader rationality forum with a major AI safety section.
• Why it stands out: Some posts are highly technical or conceptually deep, especially on alignment theory and agent behavior.
• Update frequency: Regular but less predictable than a journal.
• Best for: Conceptual and technical explorations, often with strong discussion.

4. AI Safety Newsletter / The Alignment Newsletter

• What it is: A curated newsletter summarizing recent AI safety research.
• Why it stands out: Good for tracking the field without reading every paper in full.
• Update frequency: Historically regular; check current status depending on the specific newsletter.
• Best for: Monitoring recent work efficiently.

5. Distill (archived, but historically important)

• What it is: A research publication focused on clear, technical machine learning explanations.
• Why it stands out: High-quality technical writing, including interpretability work relevant to AI safety.
• Update frequency: No longer active, but the archive is valuable.
• Best for: Deep technical explanations and interpretability fundamentals.

6. OpenAI / Anthropic / DeepMind safety research blogs and papers

• What it is: Org-specific research outputs.
• Why it stands out: Often highly technical, especially on alignment, interpretability, robustness, and RLHF-related work.
• Update frequency: Regular, though dependent on the organization.
• Best for: Applied and frontier-model safety research.

7. MIRI technical agenda posts and papers

• What it is: Research from the Machine Intelligence Research Institute.
• Why it stands out: Highly technical, often formal, especially on decision theory, agent foundations, and logical uncertainty.
• Update frequency: Less frequent than some others, but important.
• Best for: Foundational alignment theory.

If you want, I can also give you:

• a ranked list by technical depth
• a ranked list by update frequency
• or a weekly reading stack for staying current in AI safety research.