Memoria 2025 was a one-day, ~120-person unconference/festival on spaced repetition, incremental reading, and memory systems, held at Lighthaven in Berkeley, CA, on September 22, 2025, and organized by Saul Munn and Raj Thimmiah.

The goal of this post is to provide context on the event — what caused it to exist, what happened during the event, etc. I (Saul Munn) wrote this from my own perspective.

Our goals going into the event

Roughly ordered, by priority:

1. Throw down a bat-signal for people with years of experience using and tinkering with memory systems like Anki, Supermemo, Remnote, Mochi, etc, to meet other folks who were attracted by the same bat-signal. Foster durable connections, start long-term friendships & partnerships, etc.
2. Get a more & better projects started in the space.*
3. Push the frontier of memory systems usage forward.*
4. Raj & I enjoy ourselves fun.

*Within 2 & 3, Raj was especially-but-not-exclusively interested in incremental reading projects/frontiers.

But on a deeper level, I just really wanted to attend something like Memoria — and it seemed to me that the only way that was going to happen was if I organized it.

What went well

• Lot of people: ~120 tickets purchased, ~120 total attendees (attrition from ticket-buyers cancelled out folks to whom we sent free tickets, like guests of honor), ~100 people max concurrently
  • (Side-note, but basically every major event that I run, I’m shocked at how many people register at the absolute last minute. We got ~half of our tickets in the four days before the event. Nuts!)

• Seemingly high proportion of the memory systems community — at least, many of the people I was hoping could make it ended up coming
• Incredible density/signal of people
  • Raj & I were aiming for Memoria to be more of a place for experienced, non-beginner memory systems users to come together. I think we basically succeeded — e.g. Michael Nielsen had planned for his opening talk to be something like “bringing everyone up to speed,” but when he did a raise-your-hand poll at the beginning of his talk, such a majority of people already had so much experience with memory systems that he just pivoted his whole talk to an extended Q&A (which I think went wonderfully).
  • Also many people from adjacent areas, e.g. neuroscience
• On the whole, attendees enjoyed the event (more analysis at a later section):

• The sessions were pretty great. Some examples of talks/workshops:
  • 10 Important (and Underappreciated) Observations, with Michael Nielsen
  • Augmenting Attention, with Andy Matuschak & Taylor Rogalski describing their in-development app
  • Deliberate Practice for Software Engineers, with Sameer Ismail
  • Why can LLMs win the IMIO but can’t write good flashcards?, with Ozzie Kirkby and Andy Matuschak
  • Memory Systems need Genuine Interest, with Soren Bjornstad
  • SuperMemo Team Q&A, with Piotr Wozniak and the rest of the SuperMemo team
  • Homemade Tools Show & Tell, hosted by Raj Thimmiah
  • And also some smaller side-sessions:
    • What is worth remembering?, with Karson Elmgren
    • Are short-term memory systems possible?, with Alexis Gallagher
    • Content-Aware Spaced Repetition, with Giacomo Randazzo
    • Fermi Estimation with Spaced Repetition, with Drake Thomas
    • Anki for Math, with Lydia Nottingham
    • Making Memory Systems Widespread, with yours truly (Saul Munn)
  • (We’ll have a videos up on YouTube of most of the talks soon — will update the text of this post when they’re here!)
• Net profitable (!)
  • ~$9k in total costs, ~$12k in revenue, ~$3k in profits
  • We intend to plow all of the profits back into events like Memoria in the future. (If you’re interested in drawing from this war chest to host meetups in your city, let me know!)
• No ghastly operational failures (e.g. a literal fire a publicly-visible, largely disruptive, literal fire).
• Raj & I had a good time! In particular, we were both a bit worried about Memoria eating up all of our time/energy, and I don’t think this happened — we ended up at a pretty great 80/20.

What went poorly

We didn’t get all of the community together

• In particular, many people who would’ve been a great fit for the event didn’t come because they didn’t hear about it in time.
• Also, the gender & race ratios were extremely unbalanced. It’s unclear to what extent Memoria is representative of the underlying demographics of the memory systems community writ large, or if there was some further selection effect. Regardless, my general take on this sort of thing is: if people who’re otherwise a good fit decide not to join because they don’t feel comfortable coming, that’s bad and obligation of mine as the organizer to resolve (either because there’s some actual bad thing that could happen at the event, or because their expectations of discomfort at the event are not tracking what’s actually going to happen). This is true for issues of gender and race, but also for e.g. generic imposter syndrome, among others.

(Note the reasonably small n in the data above.)

We didn’t cause enough tacit knowledge transfer to happen

• I think the main advantage of in-person events (relative to e.g. blog posts, email lists, or videocalls) is that you can ask someone to pull out their laptop and go through their decks/cards/whatevers in front of you, no really could you please just show me the deck you’re talking about and walk through how it applies to your life, right now, yes while we’re talking — and I don’t think we emphasized this nearly enough, neither relative to talks nor even relative to casual conversations, during which people by-default do not feel comfortable enough to cause the sort of tacit knowledge transfer that is best suited to this sort of event.
• For examples:
  • We had a “homemade tools show and tell” where people signed up to give a 3-minute slotted talk about some hacky tool they’d built for their own workflow. This was awesome! One of my (& many others’) highlights of the event — it was great to see what pain points different people had in their memory systems usage patterns, and how they went about solving them. So much tacit knowledge.
  • We didn’t (but probably should’ve) hosted a session like this:
    • Everyone finds a partner
    • For 10m, partner A walks partner B through their memory system (with their laptop out, etc). Partner B asks questions, interrogates partner A about what’s going on, etc
    • After 10m, swap roles
    • After 10m, the organizer rings a bell, everyone finds a new partner
    • Repeat until evenly baked the time is up, or until there are no more pairs to match up
  • I think there are other sessions like that example that could’ve worked great toward the goal of causing more tacit knowledge transfer to happen.
  • This seems like the main failure which prevented Memoria from pushing the frontier of memory systems usage much, much further.

Not enough projects were started

• This is pretty hard to judge: feedback loops are slow and noisy with the most important parts of events (e.g. long-term friendships/partnerships, high quality startups/research projects, etc).
• My guess is that a success world looks like one in which, at the end of the event, 3-8 projects were on my radar as being in-the-works as a direct consequence of Memoria, but IIRC there's currently 0-2 of these.

We were net-profitable, but it seems likely that we could have been net-much-more-profitable, by maybe half an OOM

• A few clarifications:
  • Neither Raj nor I were working on Memoria for money — indeed, we both went in expecting to lose a few thousand dollars. I go into more depth on the economics later in this doc, but — I’m
  • Lighthaven gave us a huge discount, as (a) we were extremely flexible about dates, (b) Raj & I had already run many, many successful events at Lighthaven, such Lighthaven staff already had a pretty great sense of what it’d be like to work with us & that we’d need significantly less support than the median Lighthaven-event-organizer; and, (c) they were independently interested in supporting Memoria (indeed, we gave Lightcone a sponsorship to Memoria in thanks for this support).
• The main screen/monitor for the main talks half-broke at lunchtime: a quarter of the screen went black; after a few hours, this was fixed, but the colors were way wrong and the content was quite blurry. These issues caused us a decent amount of operational hassle during the event (to physically move talks around, to spend volunteer-hours trying to fix this, etc), then again caused some of the talks to be worse before we realized that the coloring was off. To be clear here, I’m quite grateful for Lighthaven’s help in trouble-shooting this — IIRC at different points three different members of their staff independently tried their hand at fixing this, but the problem was that the socket that the screen had been plugged into blew up.
• We had a scare with the venue ~2-3 weeks before the event, which had (at the time) a ~30% chance of causing Memoria to need to completely switch venues. (Mox graciously offered to host us in the case that Lighthaven didn’t work out — incredibly grateful for this.) Everything ended up being basically fine, but these shenanigans still caused me & Raj quite a bit of stress ex post, and could have made the event quite a bit worse.

Stats from the feedback form

Some fun charts & graphs — made by Claude & not fact-checked against the original data, so take them with a grain of salt:

Misc

• YouTube videos: we’ll publish the recordings of the talks soon. Thanks for your patience! There will be a link here, but if you want to know when they’re published, I’ll send out a tweet.
• Future plans: we’re still figuring out what the future of Memoria should be: if we should run another Memoria, what it should look like, etc. If you have thoughts or just want to chat — especially if you’ve already made it this far into the post — let me know!
• You can read through some of the kind words that people wrote about Memoria here: Kind words people have said about Memoria

Thank-yous

Sponsors:

• Mox, a coworking & events space in San Francisco
• Lightcone, a non-profit that builds tech, infrastructure, and community to help humanity’s future

Lighthaven, for hosting us at their lovely venue.

The folks who bought supporter-tier tickets — tickets which cost >4x the price for zero benefit to themselves: Stephen; S.; Michael; and, Andy.

Our 6-8 friends who volunteered their day to make the event run smoothly: Nat Kozak, Jeremy Rich, Jingyi Wang, Manu, and Sid Bagga.

Most of all: thanks to Raj Thimmiah, my co-organizer. This was the first time I’d worked closely with Raj, and wow I am so incredibly glad that I brought him on. Raj is incredibly competent, relentlessly resourceful, and — perhaps rarest of all — a fun guy to hang out with. Memoria would have been substantially worse (and might not have happened at all!) had Raj not stepped up to help. I’m grateful for his work, and look forward to the next chance I can get to continue working with him.

Appendix: a history & motivations

A history of the vision & logistics

In November 2024, a few months after co-organizing Manifest 2024, I drafted a doc titled “SRS-tival,” outlining some high-level details on a “conference/festival celebrating spaced repetition and other tools for thought.”

A month later, at Minifest, I ran a small meetup for SRS; 6 months after that, at LessOnline, I ran a second one; the week after that, at Manifest 2025, I ran a third one. I used these to validate demand: if nobody who came to events on adjacent niches (blogging/rationality/world-modelling, prediction markets/forecasting) was also interested in meetups on spaced repetition, then a whole dedicated event to SRS would probably flop. But the meetups went pretty well — each was larger and a better attendee experience than the one before it. This also gave me the opportunity to iteratively experiment on the format, which led me to a better understanding of what a good, bigger gathering with the same sort of people might feel like.

Back in February 2025, the Lighthaven team had kindly & very-tentatively set aside a weekend in November 2025 for me to run a fairly ambitious, 250–400-person, 2.5-day memory systems unconference at Lighthaven. But when, in mid-summer of 2025, Lighthaven noticed that they had some open weekends near the end of September (and some other events vying for early-November), they offered me a fairly deep discount in exchange for moving the date, I agreed.

That, however, meant that I had about five weeks from “begin seriously organizing the event” to “today’s the day.” Five weeks is really not that much time to take an event from a two-page Notion doc to reality all by yourself, so I asked Raj Thimmiah to organize the event alongside me.

Raj had quite a lot of experience using SuperMemo, and quite a lot of experience organizing logistics/operations for events. I’d later learn that asking Raj to join was probably the single best decision I’d make in all of my Memoria-organizing.

For those five weeks, Raj and I spent 10-20 hours each week working on Memoria in our spare time — Raj had a full-time job, I was a full-time student. And in just those five weeks, we successfully pulled off the first event that pulled together the memory systems community.

Why Memoria?

Over the course of 2024, I started using memory systems more deeply. But when I looked at what the surrounding community, I noticed a fragmented, almost-entirely-online set of subreddits and groupchats and Discords. I wanted to connect with other memory systems practitioners, but basically couldn’t.

Just as these facts were crystalizing, I read Michael Nielsen’s excellent essay “How to make memory systems widespread?” One line stuck out to me:

What will it look like to grow a memory culture?

It'll look like lots of little informal communities of practice, […] people nerding out, trying to push the edge of their art. […] Better and deeper communities will form […] people will run memory conferences to develop the practice.

I noticed that I was an event organizer, and that — relative to basically anyone else in the memory systems community — I was in a much better position to throw something like Memoria.

So I did.