Epistemic Status: Didn't think through exactly how I worded things.

tldr: When you have insightful conversation, write it down and share it so people can build on it (instead of just sharing in person). Most of humanity's power comes from being able to build complex thoughts out of other thoughts and transmit them across the world.

This is a rehash/re-examining post. Related:

This is part 2 of N of my "Ray writes down conversations he had with people" series. It's also the most adorably meta of them.

A month ago I was talking with Oliver Habryka about why Less Wrong was important. (Call us biased if you will). One thing we both noted: in the days of yore, it seemed like a lot of prominent scholars/thinkers wrote down their insights and research on Less Wrong. Then, eventually they turned professional and joined official organizations whose job was to think fulltime.

Also over the past few years, those organizations (including but not limited to MIRI, CFAR, Givewell/OpenPhil) shifted from being younger-with-nothing-to-lose to older-with-reputations-to-safeguard, and their public facing tone seems to have shifted from "earnestly sharing thoughts as they come up" to "carefully crafted PR statements."

Thirdly, a lot of people moved to major geographic hubs, where it became easier to have in person conversations than to communicate via written blogpost. So... that's what people have tended to do.

I sympathize with the notion that people are busy and writing things up is a) time consuming and b) potentially risky. But I think the consequences of this are at least underweighted.

At least, I think people having informal conversations should make more of an effort to write those up in accessible form when there aren't reasons to be cautious.

Consequences of not writing stuff down include:

  1. If you're not plugged into the personal-conversation-network, it's hard to keep up with a lot of collective insights relating to both rationality, effective altruism and x-risk. (This results in weird filtering effects which aren't the worst - being able to network your way is a credible signal of something. But I don't think it's the best possible filter)
  2. It's actually fairly time consuming to propogate ideas via in-person conversation. Like, one of the major advantages of humanity is ideas being able to efficiently spread via writing.
  3. Another major advantage of writing is being able to increase your collective working memory and build on ideas. When much of your insights are developed via conversation, not only are you preventing far-away-people from building on your ideas, but you are hampering the ability of yourself and your immediate colleagues to build on those ideas. Writing things down (and then turning them into essays with titles that can be easily referenced) makes it easier to build complex models.

I think people feel a lot of pressure to write things up well, and as a result don't write things up at all. So my current take is, if you have a conversation that seems to contain important insights, err on the side of getting it written up quickly without as much regard for being timeless.

Later on if you think it deserves to be written up in a more timeless, scholarly form, you (or someone with more time) can still do that.

Moderation Guidelines: Norm Enforcing - I try to enforce particular rules (see below)expand_more

I suspect that a big problem is people thinking, "That conversation did feel kinda insightful to me, but it probably won't be inightful to others."

Suppose that is a problem. How can it be dealt with?

  • It would be nice if there was a quick, informal way to gauge whether there's interest in a write-up. Perhaps a Slack channel could be used for that?
  • It may be a good idea to err on the side of caution and write the conversation down anyway.

Being outside of the valley, I definitely feel that there are conversations happening there and things generally known there that are not perolating their way down to us. I would love for an insider to volunteer to do a monthly write up on the kinds of in person discussions that are happening in the valley.

The most obvious way to write down a conversation is to record and transcribe it. We've tried recording meetups in the past and this reliably fails to result in much, presumably because they aren't transcribed. I presume that the technology has reached a point where record+transcribe is reasonable. What is the right tech to do this, presumably using a smart phone so it doesn't require carrying around an extra device?

I predict this, at best, results in something lower effort but less good than a post summarizing the relevant bits. There’s a lot of weird noise in the actual conversation (pauses, occasional yelling, weird references only the two of you get, rambling, repeating yourself) that made sense at the time but doesn’t make for a good reading experience.

Also, you often won‘t know you’re having an important conversation till the very end, and part of the point here is to capture the value of things that are happening spontaneously.

There's "Rev", which transcribes at 1 dollar a minute. I've used this in the past to write blog posts from conversations, but it doesn't seem worth it for something like an hour long meetup.

Could you say more about what kinds of things you think should be written up, ideally with examples? This feels too broad to me.

The previous example (other than this post itself) of "Ray basically writes up a conversation directly into a blogpost" was Guarding Slack vs Substance.

Others coming up will be a couple conversations with Andrew Critch, "Using fiction to train model building" and "Turning Money into AI Understanding".

This isn't actually much different than the more general concept of "notice interesting ideas to write about and write about them", but it's a bit easier because you don't need to come up with all the ideas and refine them yourself, you can just basically write up the conversation. (Sometimes it turns out to be structured more blogpost-like).

The title "Using fiction to train model building" makes Critch's suggestion make so much more sense. When I saw the original formation of it that you gave, my reaction was "why would I do such a thing?" and now that you say 'to train model building' that makes perfect sense. Note that you can also use the actual model of what caused the fiction to look like it does to model build...

One possible solution: erring on the side of having conversations online.

This could be taken to a stronger extreme than "erring on the side of" - two people hanging out in person could just text message each other instead of talking. That does sound extreme, but perhaps it could be appropriate in some scenarios. One benefit is that, by using something like LW, you could utilize threads and thus better keep track of things in the conversation. Sometimes when I'm having an in person conversation I'll open up a Google Doc and create an outline of the conversation for both conversation members to look at.

I personally find conversations in person to be much more efficient and positive valence, such that if I talk to someone occasionally in real life and often on the internet I tend to save topics I think will be relatively important / high value for in person discussion.

In contrast I find internet conversation is often more conducive to fishing for high value ideas, since there's so much more input.

This isn't to say that high value conversations can't happen on the internet or that that style might not be sufficient for some people, but I don't think I could make texting someone next to me work for me. As an example, in past experience I've found myself crossing the office to make / elaborate on requests initially made over chat.