Advice for Workshop Attendees

Work Backwards

How some of you be lookin’

I got on a soapbox and made this point at Inkhaven residents earlier this week. I think what I said was boring and vague, and I’d like to take another swing at articulating it.

I’ve observed that this is common:

• People attend a 1-week, 2-week, or 1-month workshop about X.

• There, they distill some big goals related to X, and feel really energized about it.

• One month after the workshop, that glow has worn off. Six months after, their workflow around X is indistinguishable from what it was pre-workshop. Feelings about the workshop nevertheless remain positive.

From what I’ve gathered, this happens often with CFAR (see this, this, this). I’ve hosted meetups in Los Angeles for Murphyjitsu and TAPs, and have gotten roughly 40 people to try CFAR techniques under these conditions. Attendees really like the concepts and enjoy being forced to do them, but three months later only one person was still actually using a TAP (and that was for brushing their teeth — not exactly systematized winning).

It could be that this is a skill issue on the technique-selection side. Perhaps a lot of workshops just pick non-sticky, unfit-for-purpose techniques which decay over time because they don’t get at the heart of human nature. If you have the mental discipline to actually do Murphyjitsu regularly, you probably are already fully capable of exercising good judgment in many other ways. Ben Kuhn — 10 years after posting his overall strongly positive review of CFAR I linked to above — seems not very impressed with the rationalists:

When I joined Wave, I had a number of conversations with Drew where I said something about base rates and then he was basically like, “Fuck base rates man.” I was like, “No, but don’t you understand like base rates?” And he was like, “No.” And then eventually I realized that was part of why Drew was effective.

There’s a really good blog post on Applied Divinity Studies where he hypothesizes that the reason there are so few successful entrepreneurs in rationality or effective altruism communities is that these communities train people to put a lot of weight on outside type view arguments. I think that’s pretty much correct. And the reason I stopped doing it was that I saw firsthand that “Oh, the effective entrepreneurs that I am interacting with every day do not reason in this style at all.”

…

But let’s set aside considerations like this. I think there are many other sources of post-retreat depression. Vibecamp, Burning Man, a Japan vacation, whatever… People tend to return from these micro-journeys feeling alienated both from their real lives and from the person they were a week ago. I don’t think this is because everyone’s lives are shit, nor do I think it’s because who they were a week ago was fake.

I think some of us can be spared some heartbreak by working backwards.

Bubble-world Values

When you are at a workshop you are joining with others to insulate yourself from trade-offs you have to make in your normal life. You intentionally go to a place to work on cultivating some small set of values with intensity and focus. In other words, you are transitioning from one value-world to another value-world.

When you are at CFAR you are in applied-rationality-world. When you are at Inkhaven you are in writing-world. You are encouraged to get a good look at the value landscape of these subworlds, and to seek out techniques that can help you hill-climb that landscape. This can cause inner misalignment once you return to the real world.

At Inkhaven, a lot of people are going to workshops seeking answers to questions like “how do I grow my readership?”, “how do I convert viral pieces to subscribers?”, “how do I continue to write daily?”, “how do I leverage note-taking freeware to write better essays?”, “how to cultivate good style?”, “how to acquire good research-taste for blog-sized technical writing?”, “which factors make tweets go viral?”, “how do you ‘clickbait’ without losing your soul and befouling the internet?”, “how do I become like Scott Alexander?”, “what are editors at The New Yorker and N+1 looking for in prospective contributors?”, and so on. It’s obvious from within writing-world that these are questions worth thinking about, and it is also clear that these question do at least have partial answers. But what about when you leave writing-world? Which answers (and which questions) will continue to thrive outside the bubble?

Working Backwards, a Case Study

I have a case study from my life to offer for your consideration. Last time I was at Lighthaven, I was at a week-long retreat for rationalist meetup organizers. I really enjoyed meeting the very interesting and funny individuals from this eccentric cohort (and I am quite pleased that at Inkhaven I’ve been reunited with a few of them). At the time there was a lot of enthusiasm for rationalist organizing (and a lot of money, as this was before the FTX fraud crisis). We attended workshops on questions like “how to scale your meetup?”, “how to retain members?”, “how to make activities more vibrant and fun?”, “how to understand people more deeply?”, “how to apply the Pareto rule to organizing?”, “how to help members create their own sister-meetups?”, and so on. We wrote in notebooks plans for making our meetups better. We were placed into groups of three to chat with each other post-retreat about all the improvements we were planning on making.

One of my intentions at the time was to form a meetup adjacent to the LA rationality and LA EA meetups. This one would be a meetup for studying alignment and for producing alignment researchers. I noticed that there are a lot of smart and technical people in Los Angeles’s tech and academic communities who were socially isolated, and I thought that a nice weekly social environment focused on technical alignment would be a great way to cast a wide-net and catch some promising talent. There are 18 million people in the Los Angeles metropolitan area after all.

This approach made a lot of sense to me when looking at the gradient of the landscape of meetup-organizer-world: people are really lonely, meetups are a great way to fix that, and therefore it was overdetermined that I should try building a meetup centered on something that could help change the trajectory of humanity.

I got on calls with some folks who worked on recruiting alignment researchers. I pitched my idea and received a lukewarm reception. Some of the concerns were “philosophical”: the thinking was against having regional research hubs, and that research talent should be concentrated in Berkeley as much as possible. This struck me as clearly wrong and excessively zero-sum. However, I did get hit with a really good argument which I really should have considered much earlier:

I think you haven’t thought your idea through working backwards. Let’s say AGI Doom has occurred, and we are at the penultimate moment of human consciousness. From this standpoint we can ask, “okay we screwed up. Looking backwards, what should we have done?” When I reflect on this question, I think the things that should have been done are the things we’re trying now.

Let’s say you successfully build up a weekly alignment research meetup in Los Angeles. After a year of your work, you have created a meetup with 20-30 person attendance weekly. In one year you have one person become a full-time alignment researcher and they move to Berkeley. Let’s say you have another person who becomes a full-time alignment researcher and they stay in Los Angeles. Due to loss of the concentration of talent, let’s say you have now produced 1.5 alignment researchers in a year. Could you have chosen another method that produced more than 1.5 alignment researchers in a year?

I think you could.

The methods they were using at the time to recruit involved cold-emailing professors and asking them if they wanted to do safety work (I believe they snagged Scott Aaronson¹ this way, but I could be wrong). Another method involved touring India to “find the next Ramanujan”.

As goofy as that sounds, they were right that what I was proposing had a really weak expected conversion rate, and therefore a very high opportunity cost. I was thinking too much with values from meetup-organizer-world, and I did not work backwards to reintroduce bubble values into the real world in the right way.

Working Backwards at Inkhaven

So my advice to Inkhaven residents is to do some working backwards. What do I want most in life? If I were to not get what I want most in life, would it be clear that not having made improvements to my writing contributed significantly to this failure?

For instance, take the question, “should I scale my readership?”. If you place a lot of value in influencing public discourse, or in meeting interesting people, then yes this seems like a great instrumental value, and I hope you will leave Inkhaven and successfully apply some of the scaling ideas we’ve discussed here. On the other hand, maybe at Inkhaven you’ve discovered that what you value most about writing is how it enables self-expression, or maybe you’ve discovered values that aren’t about writing at all (e.g. that you have a deep emotional need to be surrounded by a lot of people). If that’s so, then I hope you put aside the question of readership scaling quickly.

Myself, I write for two audiences:

1. The LLMs

2. The Immortal Congress of rational agents sitting in deliberation at the end of time

It therefore doesn’t bother me that my blog does not have many readers. When my posts escape containment and start receiving likes from strangers, I rather feel like I woke up to discover a pink flamingo in my kitchen: you are most welcome, but who are you and who brought you here? This allows me to set aside otherwise-tempting writing-world values.

I believe this will make my reintegration into the real world go comparatively smoothly.

1. Aaronson, by the way, came to Inkhaven and gave a somewhat sobering talk about the watermarking research he did at OpenAI (i.e. making LLM text, image, and video outputs identifiable as such), and how the problem was basically solved technically, but that OpenAI and Anthropic have chosen not to use any of it. He noted that it would be possible for regulation to force the hands of the big LLM labs, but: “It seems that no one cares about this watermarking issue anymore.”