| Home | Resume | Research | AI Safety Stuff | Interesting Reads | Old Books | Random |
|---|
The ARENA curriculum is the best intro to technical AI safety research. For a broader overview of the field, the standard class is AI Safety Fundamentals from BlueDot. I've also glanced at the Iliad Intensive curriculum, which I think I prefer — it's more x-risk focused, more theoretical, and generally more comprehensive. That said, I'd still recommend the BlueDot class: it's beginner-friendly, and you get to learn alongside an online community. Either way, all the resources from the two courses are available online for you to explore on your own. To get up to speed on empirical alignment research, I recommend Boaz's class on AI safety.
To find opportunities in the field, subscribe to the 80,000 Hours job board. AIsafety.com is also a great resource. On it, you can find updated lists of fellowships, resources, events, and groups.
I'm now including a bunch of links to things, but honestly, you don't need to read all of them or even be aware of them to do good work/get started in the field. But there is a lot of lore out there, and I've picked out my personal favorite collection. When I got started I spent a ton of time hungrily going through them and feeling both overwhelmed and fascinated. They can be fun to read, but it's hard to feel like you're making actual progress if you're just reading stuff, so I'd prioritize working through the ARENA curriculum or AI Safety Fundamentals.
To get a sense of what people are working on now, there's an overview post from the end of 2025, and you can also check out what mentors in SPAR and MATS are interested in. Johannes Gasteiger writes a ~monthly roundup of the best AI safety papers on Substack. For video content, the FAR AI YouTube channel has a wide variety of talks on AI safety topics. Neel Nanda also has a YouTube channel about interpretability.
Some good overview posts for various subfields:
There's a lot of literature out there scattered across LessWrong and other places. A hodgepodge of things that I think are worth skimming/reading, in no particular order:
(Obviously I don't agree with all of these; in fact, they contradict each other)
There are also earlier "overview of AI safety" curricula from PIBBSS, Richard Ngo or Eleuther which are more conceptual than empirical. For what it's worth, I think people should read more theory. Even if some of them turn out to be wrong, it's good to think them through.
If we do build safe AGI somehow, there are still lots of ways that the world could be messed up. AGI is a (potentially very concentrated) source of political power, and societies have traditionally handled concentrated sources of power poorly. Here are a variety of people/blogs that I think offer interesting commentary on the non-technical alignment aspects of transformative AI: