AGI Safety resources

I manage BlueDot Impact’s AGI safety fundamentals website; all of my AGI safety-related recommendations can be found on there. It has career advice, opportuinties and other resources for getting involved with AGI safety work. You can contact the team if you feel any advice that would be useful is missing.

App recommendations

Here is an up-to-date list of the apps and services I use and recommend.

Contacts & meetings tracker (Notion template)

Since I started having more meetings, I have needed a way to keep track of them all. I can now quickly see all the meetings I’ve had with a person, and all the people I know at an organisation, etc.

My template has more details and instructions on how to get set up in 15 minutes.

Book summaries

I write summaries of some of the books I read. When I think a book is great and have made legible enough notes, I make them public. I tend to read about building organisations and doing good product development, but may add other (non-fiction) topics here in the future.

Book summaries database.

Getting started in Python and Machine learning

This is a guide I wrote with my advice for learning python and machine learning (theory, and practically in python) for the first time.

Newsletter recommendations

These are some newsletters I follow and recommend.

  • Jack Clark’s weekly Artificial Intelligence Roundup – good for keeping up with latest developments in AI. A slight gearing towards people thinking about governance of AI, but covers technical advances for general interest.
  • Chartr – a weekly email displaying 3 stories or other current affairs with data. I appreciate gaining a quick intuition for the scale of phenomena, and remember these much better than words in a news article!

Talks I have given (which you can use)

I have listed the talks I have given on this page.

If they will be useful to you, you are welcome to just use the slides. If you do use the slides, I’d appreciate: 1) accreditation, 2) you let me know that you’re going to do so, 3) whilst letting me know, let me know if you have any feedback, whether you changed anything, and how your audience responded!

  • ‘Intro to AI safety (why worry?)’. This talk is intended for a beginner audience, and doesn’t assume too much ML knowledge. It is particularly targeted at an audience which is aware that some people are worried about AI safety, but they aren’t sure what the exact arguments are or why the naive solutions won’t work. It tries not to wave too many hands and to be thorough (which does make the talk longer, at ~1hr), and it motivates why these are not easy problems to solve.
  • ‘Overview of where and how you can do alignment and governance work’. This talk makes the case that there are actually careers you can pursue in AI safety. Having introduced the problem concretely with the previous talk (or if your audience is already aware of some challenges in AI safety), it’s nice to show people that there are tangible things they can aim for to have an impact. This is a more subjective talk, in terms of my take on what helps, and is more likely to fall out of date.

Other recommendations