Sometimes, people say, "Wow! given LLMs, and other recent AI develpments, it looks like we're at the start of a slow takeoff."

I think that's only half right. The existence of Large Language Models _are_ evidence about how impressive AI is going to be in the RAMP UP period to AGI.

But it is 0 evidence about whether there will be a SHARP BEND in the curve, at some unknown point in the future.

Sometimes people say "Wow! given LLMs, etc, it looks like we're at the start of a slow takeoff."

But I think that's only half right. The existence of Large Language Models _are_ evidence about how impressive AI is going to be in the RAMP UP period.

But it is 0 evidence about whether there will be a SHARP BEND in the curve, at some point.

Here's a long list of lenses for making sense of the (social) world, with associated links.

As in, if you don't have this concept in your tool kit, you're missing an important driver of what makes our world tick.

I'd love it people could add to this list. What am I missing? SOCIAL CLASS (as distinct from economic class)

siderea.dreamwidth.org/1237182.html

It seems like it should be relatively cost effective to have "sleeping planes", in the style of sleeping train cars.

Instead of buying a seat, you buy a bunk. Everyone is horizontal for the whole flight. (You would still strap in for safety.) If you stacked the bunks it seems like this shouldn't result in more than a factor of 2 reduction in passenger capacity.

(Maybe it would be tricky to set up a way for people in window bunks to get out to go to the lavatory etc. That might also cost some real estate.)

This is a great post. There's something about it a lot of the magic spark of my favorite classic SSC posts.

It makes me feel pride and I want to signal boost it.

astralcodexten.substack.com/p/trying-again… I think what I love about it is grappling with the hard problems of knowing what's real and how to figure out what's real.

And that it makes a case for "Self-ing irrationality", as Scott once put it.

slatestarcodex.com/2014/04/15/the….

Has anyone done an improv workshop in which you did the exercises described in Impro?

I'd like to do one, ideally with an experienced facilitator. (I'm aware that Keith Johnstone still mentions workshops on his website, but there hasn't been one in a couple of years at least.)

Math question:

My experience of learning calculus was of memorizing a bunch of specific, discrete rules for finding derivatives (or taking integrals) and proving those rules via largely-tedious-feeling algebraic proofs. Plus doing tedious algebraic practice problems. My experience of learning Linear Algebra was more like taking some hard to grasp conceptual abstractions and working them over in my mind until they "fit" in working memory (and felt solid instead of slippery), or until I could see why some structure HAD to have some property.

I have the experience (I don't know if it's imagined) that my day-to-day consumer experience has been getting worse over the past decade, after peaking around 2015. This might just be confabulation. There was a general feeling of optimism and hope in my life and my world then, when I had just moved to the Bay and EA was just starting to take off.

Things felt exciting and and alive.

Does anyone know of vegan / plant based meat-substitutes that don't include canola oil, safflower oil, rapeseed oil, soybean oil, corn oil, or other omega-6 polyunsaturated fats? Dr. Praeger's Classic Chick’n Tenders was a staple of mine, but they seem to be unavailable at stores these days.

And in any case, they contain sunflower oil.

drpraegers.com/our-food/class…

Vegans, I'm going to start supplementing Omega-3 /fish oil.

It seems like fish oil can come from a number of different kinds of fish. Is consuming some more humane than consuming others? Either because the fish have a better time, or because they're less likely to be conscious? For instance, I _guess_ that:

krill > pollock

(on the theory that, the bigger the animal, the bigger the brain, the more likely it is to to conscious morally-relevant experience)

I'm finding @vgr to be exceptionally insightful and fecund for my own thinking.

I often don't get very far into reading a post before I've spun off writing an essay on some tangentially related point.

What are your personal favorite of his essays? One thing about @vgr is that he has among the best counterpoints to the the style of LessWrong rationality that I was enculturated by.

He genuinely does understand some things that Yudkowsky rationalists, and Yudkowsky rationality, doesn't see very well, I think.

I'm looking for two tweets. Maybe people can help me find them.

One from @s_r_constantin, describing how, at first, there was something hard to grasp about the point of EA, because more efficient charity means that you're sacrificing less, and therefore less morally good. @s_r_constantin One from @VitalikButerin. Its a cartoon with some grey stick figures discussing policy and deciding that more X is good on the margin.

And then one of them goes to a blue Yay-X convention, and brings up some second order effects, to predictable response.

Here's a MUCH more radical twitter feature to improve discourse. This one could change the world.

*Signal boost tweets in proportion to forecasting accuracy.* This idea was inspired by / originated with @KelseyTuoc. Some of the specific implementation details below are my own elaboration, but credit goes to her.

I am surprised with how unimpressive Elon's new features for twitter have been. Most seem to focus on exactly the wrong parts the platform for creating good discourse.

Here are some feature-ideas that seem likely to make the app better for the world or for discourse: 1. An open, un-rate-limited API.

Twitter is and rich, extremely valuable dataset for social science, and real-time information tracking.

It's kind of ridiculous that you can approximately only access that dataset through the twitter UX.

This is a pleasant dream. But here's what ACTUALLY happens to Spreadsheet Altruism:

Making spreadsheets about the real world is hard and annoying and your results are way WAY less solid and rigorous seeming than it seemed like they would be from the word "spreadsheet".

https://twitter.com/EpistemicHope/status/1604887378593603584

It's both difficult and scary to stick your neck out making a spreadsheet to maximize your altruism.

So most people would end up just pointing to 80K's and GiveWell's spreadsheets, to as the justification for their actions.

I REALLY like this.

For one thing, I would feel embarrassed calling myself a "Spreadsheet Altruist", if I didn't actually have at least one spreadsheet for steering my altruism.

Which is a feature, not a bug.

https://twitter.com/NathanpmYoung/status/1602399684730970112

EAs get to be part of the EA club by deferring to the EA network about which causes and actions are most impactful.

That deferral is fine, but it gets conflated with the narrative/branding of "doing the most good."

Doing what "EA" thinks is good != Doing the most good

I kind of regret making a post saying "this is a good critique of EA".

Because, as @vgr says, right in the thread, it's a problem of EVERYTHING that is attempting to do something at scale, not EA in particular. And I'm more interested in the broader point.

https://twitter.com/EpistemicHope/status/1592372594006638598

Also it is a good stab at the ideology of Effective Altruism, which tends to have as an implicit assumption that key opportunities are scalable. That if you can turn a little money or effort into something good, you then you must be able to turn a lot of money into a lot of good.

Why do we typically measure economic well-being with GDP, rather than national wealth?

It seems like GDP has a number of crucial problems which cause it to measure something other than what we intuitively care about. 1. GDP measures how much money changes hands, instead of how much value is created.

If a supply shock reduces the supply of a good by half, but drives up the price of the good by more than 2x, this is counted as a higher GDP than baseline, even though _less_ wealth was created.

I found this post (after having read the post to which this one is responding), quite compelling, in contrast so some other posts on related themes.

benjaminrosshoffman.com/parkinsons-law… Specifically, @ben_r_hoffman says "Stealing the locals' land to plant trees and raise ponies is a totally bonkers response to the three summary facts enumerated."

I paused to think for a bit about what interventions I feel inclined to try, and I agree: these ones seems crazy.

An epistemic stance: knowing and emphasizing that people very often fool themselves, and that using hard data is one way (or the only way?) to keep yourself moored to reality. Like, we know about Heuristics and Biases, and we know about confabulation. People overweight their own anecdotal experience. It is really easy to generate a theory that explains imagined data.

Individual humans have unique cognitive profiles. I'm curious if we could intentionally train machine learning systems to similarly have a "cognitive style." [thread] The most robust finding of psychometrics is the positive manifold: that scores on virtually all cognitive assessments are correlated.