How do I choose between different AI safety research publishers for learning technical safety and policy angles?

A good way to choose is to treat “AI safety publishers” as serving different learning goals, not as competing on one axis. If you want both technical safety and policy angles, you’ll usually get the best results by combining a few sources rather than picking only one.

1) First decide what you want to learn

Ask which of these matters most right now:

• Technical alignment / robustness / interpretability
• Governance / policy / standards / evaluation
• Career orientation: where to build depth and credibility
• Speed vs rigor: cutting-edge but messy, or slower but more reviewed
• Practicality: actionable frameworks vs conceptual discussion

2) Use these criteria to compare publishers

For each publisher, check:

• Audience
  • Researchers? Practitioners? Policymakers? General readers?
• Review process
  • Peer reviewed, editorial review, or self-published?
• Methodological quality
  • Empirical, theoretical, case-based, or speculative?
• Citation/impact
  • Are their ideas cited in academic, policy, or industry work?
• Balance
  • Do they cover both technical and governance questions, or only one?
• Timeliness
  • Do they respond quickly to new model developments?
• Bias and agenda
  • Are they advocacy-oriented, neutral, or strongly tied to a specific viewpoint?

3) Typical publisher types and when to use them

A. Academic journals / conference proceedings

Best for:

• Rigorous technical work
• Methodological depth
• Learning what is considered credible in research

Tradeoffs:

• Slower
• Less policy-oriented
• Often narrow in scope

Use when you want:

• Strong foundations in technical safety
• To learn how claims are justified

B. Think tanks / policy institutes

Best for:

• Governance, regulation, standards, international coordination
• Policy memos and scenario analysis
• Translating technical issues into policy questions

Tradeoffs:

• Less technical depth
• May reflect institutional priorities or policy positions

Use when you want:

• Policy literacy
• Exposure to actionable governance ideas

C. Independent research blogs / newsletters

Best for:

• Fast-moving technical ideas
• Early-stage arguments
• Cross-disciplinary discussions

Tradeoffs:

• Variable quality
• Harder to assess rigor
• Can be very opinionated

Use when you want:

• Breadth and current debates
• Leads for deeper reading

D. Dedicated AI safety organizations’ publications

Best for:

• Specialized safety topics
• Frontier-model evaluation, alignment, interpretability, risk
• A blend of technical and governance work

Tradeoffs:

• Often aligned with the org’s mission
• May emphasize particular threat models or policy preferences

Use when you want:

• A focused safety lens
• Practical research direction

4) A simple selection strategy

If you’re unsure, build a stack like this:

• 1 rigorous technical source
• 1 policy/governance source
• 1 fast-moving commentary source

That gives you:

• Depth
• Policy context
• Current debates

5) Questions to ask before trusting a publisher

• Who writes for them?
• What kinds of claims do they make?
• Do they distinguish evidence from speculation?
• Do they cite primary sources?
• Are they transparent about uncertainty?
• Do they update or retract old content when needed?

6) Practical recommendation

If you’re learning from scratch:

• Start with reviewed technical material for core concepts
• Add policy briefings and reports for governance context
• Follow a few high-quality newsletters/blogs to stay current
• Keep a note of which claims are:
  • established
  • debated
  • speculative

7) Red flags

Be cautious if a publisher:

• Never cites sources
• Uses lots of confident language with little evidence
• Treats one research agenda as obviously correct
• Confuses technical uncertainty with policy certainty
• Doesn’t separate analysis from advocacy

If you want, I can also give you a comparison checklist you can use to evaluate specific publishers side by side, or suggest a curated mix of publishers depending on whether you care more about technical safety, policy, or both.