ChatGPT Recommends

AI Safety

Asked to recommend a AI Safety reading list, ChatGPT came up with the following. Interestingly, it didn’t provide hot links to everything. A pretty obvious thing to do, but tedious so it just decided to omit that step. A perfect example of how AI works these days (“what can I get away with?”).

A Curated AI Safety Reading List

1. Daily / Weekly News

These answer:

What happened this week?

MIT Technology Review (AI)

Probably the best mainstream technology journalism covering AI.

Strengths:

Reuters AI

Probably the most reliable factual reporting.

Strengths

Excellent source for tracking chronology.

Financial Times (AI)

Excellent for:

Very useful for understanding deployment pressure.

The Economist — Artificial Intelligence

Best for

The Guardian — AI

Worth reading because it regularly publishes

It also gives visibility to viewpoints that other outlets sometimes ignore.

2. Frontier Research Organizations

These answer

What are the labs actually discovering?

Anthropic Research

Among the strongest sources for

Excellent technical summaries.

OpenAI Research

Important because many capabilities arrive here first.

Read:

Google DeepMind Publications

Especially

UK AI Security Institute (AISI)

Perhaps the most consistently careful government technical work on frontier evaluations. Increasingly indispensable for anyone interested in AI safety.

US Center for AI Standards and Innovation (CAISI)

Watch for

3. Think Tanks and Research Institutes

These answer

How are serious researchers thinking about AI risk?

AI Futures Project

Formerly Forethought.

Excellent forecasting work.

METR

(Model Evaluation & Threat Research)

Leading work on measuring dangerous capabilities.

Redwood Research

Particularly important for

Alignment Research Center (ARC)

Foundational alignment work.

Centre for the Governance of AI (GovAI)

Outstanding governance research.

Especially useful for

Future of Humanity Institute (legacy)

Although no longer active, its papers remain foundational.

Future of Life Institute

Policy advocacy.

Good interviews and reports.

RAND

Increasingly active on frontier AI governance.

Brookings

Excellent mainstream policy analysis.

OECD AI Observatory

International policy tracking.

4. Individual Commentators

These answer

How should we interpret events?

Gary Marcus https://garymarcus.substack.com

Probably the most important skeptical voice.

Strengths

Often right for different reasons than AI safety advocates. Reading him prevents tunnel vision.

Zvi Mowshowitz https://thezvi.substack.com

Probably the single best high-volume commentator.

Covers

His weekly summaries save enormous amounts of time.

Scott Alexander https://www.astralcodexten.com

Rarely writes directly about AI, but when he does it is thoughtful and unusually balanced.

Dwarkesh Patel https://www.dwarkesh.com

Not a safety advocate per se.

But probably the best long-form interviewer of frontier researchers.

Tyler Cowen https://marginalrevolution.com

Excellent for

Often offers a useful counterweight to existential-risk narratives.

Ethan Mollick https://www.oneusefulthing.org

Very good on

A reminder that most AI today is still tool-like.

Ben Thompson (Stratechery) https://stratechery.com

Excellent business analysis.

Kelsey Piper https://substack.com/@kelseytuoc

One of the best journalists covering AI safety.

Simon Willison https://simonwillison.net

Exceptional on open-source models, tooling, and practical developments.

5. Rationalist / Alignment Community

Essential reading—even if one disagrees with parts of it.

LessWrong https://www.lesswrong.com

Still the intellectual center of technical alignment discussion.

Many ideas later adopted by labs appeared here years earlier.

Alignment Newsletter (Rohin Shah) https://rohinshah.com/alignment-newsletter/

Probably the single best technical newsletter.

Weekly summaries of new papers.

AISafety.com

An excellent curated portal that itself recommends blogs, newsletters, podcasts, forums, and learning resources, making it a useful “directory of directories.”

GreaterWrong https://www.greaterwrong.com

Alternative interface for LessWrong.

6. Academic Sources

arXiv https://arxiv.org

Must-read categories:

Search terms:

Google Scholar alerts

https://scholar.google.com/scholar_alerts?view_op=list_alerts&hl=en

Useful for

SSRN

https://www.ssrn.com/ssrn/

Good for governance and legal work.

7. Governance

International AI Safety Report

Required reading each edition.

Likely the closest thing the field has to a consensus technical assessment.

OECD AI Observatory

EU AI Office

UK Government AI Security Institute

NIST AI RMF

8. Newsletters

Highly recommended.

Community recommendations consistently place the Alignment Newsletter and AI Safety Newsletter among the best ways to keep up with new safety work.

9. Podcasts

10. Deliberate Counterpoints

A good monitoring system should include voices that challenge the prevailing assumptions.

Include:

These writers often push back against overgeneralization while still taking AI seriously.

If You Only Read Ten Things

If he were willing to check only ten sources regularly, I would suggest:

  1. Reuters AI
  2. MIT Technology Review (AI)
  3. Anthropic Research
  4. UK AI Security Institute
  5. Zvi Mowshowitz
  6. Gary Marcus
  7. LessWrong
  8. Alignment Newsletter
  9. International AI Safety Report
  10. arXiv (alignment / interpretability / agents)

That combination provides a healthy mix of reporting, primary technical work, policy, commentary, skepticism, and frontier thinking.

One addition I would make for your project

Because From Tool to Actor is concerned with the transition from bounded tools to autonomous institutional actors, I would encourage you to organize your reading around five complementary lenses rather than by publication alone: