[August 2023 edit: I basically don't endorse this post as currently written anymore for a few reasons:

1. I think university groups should primarily be focused on encouraging people to learn a lot of things and becoming a venue/community for people to try to become excellent at things that the world really needs, and this will mostly look like creating exciting and welcoming environments for co-working and discussion on campus. In part this is driven by the things that I think made HAIST successful, and in part by thinking the post "University EA Groups Need Fixing" is right about what the role of retreats in the problems with EA groups.
2. I also think retreats are more costly than I realized when writing, and (relatedly) if you're going to organize a retreat or workshop or whatever, it should probably have a theory of change and driven by a target audience's area of interest and background (e.g., "early-career people interested in AI policy who haven't spent time in DC come and meet AI policy people in DC") rather than general-purpose uni group bonding.
3. I also think "retreat" is basically the wrong word for what these events are; at least the ones I've run have generally had enough subject-matter-driven content that "workshop" is a more appropriate term.
4. That said, I do still think university groups should consider doing retreats/workshops, depending on their capacity, the specific needs of their group, and the extent to which they buy the arguments for/against prioritizing them over other programs. I'm leaving a lightly edited version of the post up in case it has some useful information.]

TL;DR, doing retreats should be a top priority (maybe the top priority) for university groups due to their unique effects on personal prioritization.

In the one-on-ones I’ve had at EAGs (most recently xBoston) and elsewhere, a recurring theme in people’s personal stories is that they include something like “Yeah, I basically thought this EA cause prioritization stuff was right, but kept working on my other stuff, and then I went on a retreat and I was like ‘Oh shit, I actually should change my life about this.’” Or, at a later stage, they were planning on doing some “EA-approved” thing but the retreat significantly shifted their priorities towards working harder on more impactful work.

This, in addition to the “theoretical” reasons below, leads me to think that retreats are massively more effective than other bread-and-butter EA group programming like intro fellowships, speakers, etc. Personal journeys that might have taken months or years can happen in 1-2 days. I’ve also seen important misunderstandings about EA that have persisted through most or all of an intro fellowship be corrected at retreats.

Why do retreats work?

Retreats encourage the kind of sustained reflection, one-on-one conversations, and social network construction that actually get people to reevaluate their plans. Most other EA programming occurs in classroom-type settings where people are used to engaging with ideas intellectually but not taking them seriously as action-relevant, life-affecting things.

This is part of a broader observation that career decisions among high-achieving students are primarily identity-driven. Convincing people to do something different with their lives can mean opening them up to changing what feels like a core part of their identity. In my case, it needed to be safe for me to stop "forward-chaining" from an identity that said “I am a public policy generalist” and shift toward a mindset of “I try to do the most good, and I currently have certain ideas about the best path for me to do that.”

When done well, EA retreats bring together a bunch of people who might have been vaguely thinking “I really need to sit down and figure out some career stuff” or “I know deep down that the EA stuff is right but there’s something (like inertia, abstraction, or lack of peer support) stopping me from acting on it” and giving them a time for both internal and social deliberation that empowers them to move closer to the values they already want to live by.

Important characteristics for retreats to make this happen:

• The retreat features lots of people who are already pretty knowledgeable about EA. Not sure what the right ratio here is, but I'm pretty confident that less than half should be newbies (e.g. hasn’t completed an intro syllabus or done a solo deep-dive) as an upper bound.
• It happens at a time of year/semester where people aren’t incredibly busy — ideally first few weeks of the semester (when people’s schedules are especially in flux) or a long weekend or both. And ideally the fall semester, since people are especially open to new activities at the beginning of the year.
• People spend at least one night (and preferably two nights) at the retreat.
• The retreat includes a lot of time for one-on-ones (including walks, which typically become one-on-ones or at most groups of 3). There should also be some kind of structured, social reflection activity (like Hamming Circles after dinner and/or a round of "gratitudes" for ways other attendees improved your experience at the closing session).
• The retreat includes time and a central, default physical space for unstructured socializing (during which people are likely to talk about EA-related things anyway, provided a critical mass are already engaged EAs).
• It is very hard to run both the content and the operations of the retreat at the same time. For example: you can't run a content session while you're driving to pick up lunch. Consider deputizing or paying somebody to help run it.
• A content theme of the retreat should be that you can make a difference on these global problems (if you’re ambitious and agentic).
• Nature settings (especially visible horizons) seem to facilitate reflection and a pleasant experience.

Things that seem good but that I'm less confident about

• The retreat involves time for explicit solo reflection, e.g. a workshop where you write or map out ideas.
• The retreat includes “professional EAs.” In my experience, these do make the retreats more exciting and facilitate “vertical networking” where newer EAs can model EA careers, get plugged into subject-area networks, etc, but also have a higher cost, because it makes the retreat more time-consuming to plan and subjectively higher-stakes. Probably worth it to try to snag at least 1 per 10 attendees unless the organizing team is very time-constrained. Just showing up for a few hours is fine.
• Some people will probably flake, so slightly over-book the retreat. To reduce flaking, confirm most people's spot on the retreat at least a couple weeks in advance and send a lot of reminders in the preceding week.

Resources for making your retreat go well:

• Get in touch with Canopy Retreats (formerly EA Retreats).
• Global Challenges Project Guide to Running Retreats
• Yale Retreat Handover Doc
• EA Groups Resource Centre: Retreats and Unconferences
• Include retreat funding in your group funding request from e.g. Open Phil (and err on the side of more money, since you can return unspent funds)

Thanks to Marka Ellertson, Juan Gil, Leilani Bellamy, Nikola Jurkovic, and others for feedback.