Independent research in inspiring and scaling collective intelligence
Previously Head of Procurement at Anthropic starting 2021
Executive Director at CFAR, starting 2018; Productivity Coach in 2017; Senior Research Analyst and Operations Associate at GiveWell 2013-2017
[I’m coming back to the comments in this post now and feeling grateful for all the engagement with our podcast. Apologies for being very low engagement on LW for the last decade.]
I just (re-?)noticed that I never addressed your curiosity about machine editing.
Yes, we made heavy use of an automatic transcript/audio editing tool. We can get you the name of it if desired, though I recommend reaching out to me offline if I don’t reply here sufficiently quickly to any follow-up questions you or anyone might have.
Your curiosity helps me realize that I think we should consider flagging all use of automatic tools even if they don’t make use of AI @Elizabeth (I’m not even sure if this tool does or doesn’t 😬).
One of Elizabeth and my ongoing curiosities is what the best format is for optimizing clarity/trnasparency, completeness of context, and ease of engagement (among other things). My current best guess is that we will consider publishing at least two versions of each podcast, at least once we have sufficient editorial capacity:
“Complete” unedited audio with automatically generated transcript—ideally with some cursory review for accuracy including (in my favorite world that would require more talent than I currently have) a community-transcript-editting process so that people like you could make these changes (and add more footnotes) directly as on Wikipedia. (“Complete” is in quotes ‘only’ because it feels impossible to include all context in EVN/my conversations unless we record all of our interactions, which is rough because she’s become one of my top 5 friends these days).
Easy edit for fast publishing without all the tricky bits that feel like they need more context to publish in a non confusing way.
Pithy version to optimize productive engagement (without sacrificing integrity, obvi). Ideally makes heavy use of footnotes and links that link everywhere including to the relevant portion of the full transcript. If we had this I imagine it should be the most visible/clickable version.
Crowdsourced Dank EA Memes tournament based on clips? (See caption contests for a fun version of this.)
Other (please suggest)
[Additional procedural comment: In service of my own learning and increasing my engagement with LW, I’m going to try to “lean in” to writing/publishing fast at the risk of falling on my face a couple times. Thank you for your patience.]
Awesome, thank you! I’m not sure if we’re going to correct this; it’s a pain in the butt to fix, especially in the YouTube version, and Elizabeth (who has been doing all the editing herself) is sick right now.
The group I played with (same as Mark Xu’s group from comment above) decided that “S2 counting is illegal (you have to let your gut ‘feel’ the right amount of time)” and “repeating some elaborate ritual that takes the same amount of time before your card is due is illegal” (e.g. you can stick your hand 10% of the way towards the pile when the number’s 10 off from your card, and 50% of the way when it’s 5 off.)
Metaphors We Live By by George Lakoff — Totally changed the way I think about language and metaphor and frames when I read it in college. Helped me understand that there are important kinds of knowledge that aren’t explicit.
What I get from Duncan’s FB post is (1) an attempt to disentangle his reputation from CFAR’s after he leaves, (2) a prediction that things will change due to his departure, and (3) an expression of frustration that more of his knowledge than necessary will be lost.
It’s a totally reasonable choice.
At the time I first saw Duncan’s post I was more worried about big changes to our workshops from losing Duncan than I have observed since then. A year later I think the change is actually less than one would expect from reading Duncan’s post alone. That doesn’t speak to the cost of not having Duncan—since filling in for his absence means we have less attention to spend on other things, and I believe some things Duncan brought have not been replaced.
I am also sad about this, and believe that I was the person best positioned to have caused a better outcome (smaller loss of Duncan’s knowledge and values). In other words I think Duncan’s frustration is not only understandable, but also pointing at a true thing.
All of these answers so far (Luke, Adam, Duncan) resonate for me.
I want to make sure I’m hearing you right though, Duncan. Putting aside the ‘yes’ or ‘no’ of the original question, do the scenes/experiences that Luke and Adam describe match what you remember from when you were here?
Agreed I wouldn’t take the ratanon post too seriously. For another example, I know from living with Dario that his motives do not resemble those ascribed to him in that post.
+1 (I’m the Executive Director of CFAR)
What do you recommend if good data is too costly to collect?
I think that if someone has made a claim but failed to use good data or an empirical model, it should not require good data or an empirical model to convince that person that they were wrong. Great if you have it, but I’m not going to ignore an argument just because it fails to use a model.
I agree with your definitions of the two curves, although I don’t know what point you’re making by the distinction.
In either case we can ask, “how much will changes in demand affect equilibrium quantity?” In a constant-cost industry, the answer will be 1:1 in the long-run (as indicated by a flat, or infinitely elastic long-run supply curve), but as you gradually shorten the scope over which you’re looking at the market, making it a shorter- and shorter-run supply curve, it will steepen (elasticity decrease) such that the answer is “less than 1:1″.
Great! This is the only ‘complete’ argument I’ve seen that our prior for animal products industries should be that they are increasing-cost rather than constant-cost. I’m not as confident as you seem to be, but that’s more of a quibble at this point, and I’m glad we agree on the meta-prior!
The challenge then is to convince Norwood and Lusk that we want to know the long-run impact of consumer choices on animal production, not the short-run! They’re clearly estimating short-run elasticities since (a) their supply curves are way too steep, even for an increasing-cost industry, and (b) they explain their elasticities with an explanation that is irrelevant to the long-run:
Because it takes a year between the time a cow is bred and the time her calf is born, and then it also takes a long period before that cow can be transformed into beef or produce milk, it is difficult for beef and dairy producers to alter production according to changes in consumer preferences.
Given our confidence for opposing positions and your credentials my best guess is that there’s a miscommunication, in which case my guess on your correctness won’t be well defined. Perhaps there’s some critical word I’m using colloquially that has an importantly different meaning in economics.
The most important factor of production in haircuts is human labour. If you double the population of a country, then you double the number of haircuts demanded, but you also double the amount of labour supplied.
To get around the objection of increasing labor, let’s assume instead that everyone in the country decides to permanently get their hair cut twice as often as before. What do you expect the real price of haircuts to be in 10 years? Significantly higher, about the same, or significantly less? My prior is “about the same” until I have more data.
More generally (and naively), if I had to guess whether a given industry is increasing-, constant-, or decreasing-cost, two factors I would consider are: 1) How important to the cost of the product are inputs that are finite? How much of the market for those inputs goes toward making the product (as opposed to other uses that might substitute away from that input if the price increases)? 2) What economies of scale exist in the production of the product? To what extent will greater demand-per-person allow firms to approach the maximally efficient scale?
I have not seen these (or other relevant) factors used to advocate that animal product industries in particular should be assumed to be increasing-cost. (However I do appreciate your attempt to argue that our prior should be that all industries are increasing-cost in the absence of better evidence, which I argue against in a separate comment.)
Given your credentials, I can’t fathom you disagree that shifts in demand can cause shifts in the short-run supply curve in the long-run. E.g. if lots of people start/stop eating meat, the short-run supply curve will look different 5 years from now due to that change alone.
Given that I believe you agree with that, I deduce that you only mean that changes in demand cannot shift the short-run supply curve in the short-run, nor shift the long-run supply curve in the long-run. With that I do agree by definition.
Funny that the only post by someone acknowledging negative-sloping long-run supply curves AND refraining from advocating a prior that meat industries are increasing-cost (yay; that’s all I really want!) ends up sounding like a disagreement anyway due to semantics : ]
You’re the first person who disagrees with my conclusion but is willing to admit that some industries will be decreasing-cost (wherein we should expect greater than 1:1 effect on production); that is very refreshing!
in the extreme, every industry is increasing-cost, simply because of resource scarcity (consider the situation if computer game demand was so high that 90% of the population worked as game developers).
Yes, I agree in the extremely-large case. What about the extremely-small case? It’s very hard to think of a fledgling market for which the average price would be higher if the market produced 100 units instead of 1 unit. So I think that the vast majority of markets will be decreasing-cost when they’re arbitrarily small, and increasing-cost when they’re arbitrarily large. What about in between? How do we know where on the spectrum an industry is?
I agree we need a prior. If something could be positive, neutral, or negative, and I have no evidence of which it is, my default meta-prior is neutral. Obviously that is an extremely weak prior (ie I’m very open to being convinced it should be something else), but nothing in the pieces I quoted justifies an increasing-cost prior (they just discuss short-run market dynamics). Your “low hanging fruit” argument comes the closest because it’s a valid reason that points in one direction, but it is incompletely argued so far (the “in the extreme” argument is unconvincing since the same can be said of an opposing force).
The default assumption is that industries are increasing-cost, because low-hanging fruit is picked first.
I agree this effect will push towards an increasing-cost industry, but there are other effects at play that might be even more powerful such that the industry is constant- or decreasing-cost. For an extreme example, consider the market for computer games; I expect this to be a decreasing-cost industry (the more people buy computer games, the cheaper-per-quality they will be in the long-run, even ignoring technology improvements). For a more moderate example consider haircuts. How confident are you that the price of haircuts will increase the more haircuts there are in the long-run? For example, do you expect the real price of haircuts to rise dramatically as the population of a country grows dramatically? I do not.
My claim is that we shouldn’t assume that the meat industries are increasing-cost unless/until we have better reason to do so. The rationalist community pieces implicitly assume that all industries are increasing-cost (and don’t give reasoning for that example that’s relevant in the long-run), whereas the economics articles I cite show that industries can also be constant- or decreasing-cost as well.
The rationalist community, and standard economics, give the same answer, and it is only your very strange assumption (which you bury in the middle of your article!) that causes a discrepancy.
Actually my prior that industries are constant-cost is not a particularly strong one; I’d be happy if the sources I cite simply remove or justify their assumption that animal product industries in particular are increasing-cost. As I’ve shown from the standard economics citations, they are actually with me in not assuming that industries are increasing-cost (unless you’re arguing that AmosWEB and Policonomics do not represent the standard economics view).
If just one consumer decided to not buy the good at any price this is what would happen.
Yes, I’m referring to a decrease in a consumer’s purchases at any price (such as someone becoming vegetarian in a chicken market) not just at one price (in which case, if the price lowered a cent, they would re-enter the market); I agree this also counts as a decrease in demand.
The answer is that it depends, and the decrease in sales could be tiny or huge depending on elasticities.
Yes, if you mean ‘long-run elasticities’ then we agree. In fact the only thing that confuses me about your answer is what you think I should learn from an introductory microeconomics textbook.
The fall in demand will (by definition) have no impact on the supply curve since the supply curve takes into account how much firms are willing to sell at every price and so already accounts for the possibility of prices falling.
I disagree with your claim and it seems like you’re confusing short-run and long-run; I don’t think you’ll find a definition of supply curve such that it cannot react to changes in demand in the long-run. Policonomics explains such a demand-motivated change in supply (D’, S’, and E2 refer to the chart in the original post):
Once this market equilibrium is reached, one might ask: what happens if there is an increase in demand? First, this will shift market demand to D’, increasing prices because supply (for the moment) is given. This increase in prices allows for profits to be made, at point 1. However, as seen before, these profits will attract new firms, which will force supply to the right (S’), going to an equilibrium such as E2, where our firm will produce the exact same initial quantity (point 2).
Based on the comments to this article I realized a stronger appeal to established economic theory would probably be more convincing. I’ve made that appeal in a separate post: http://lesswrong.com/r/discussion/lw/lj0/misapplied_economics_and_overwrought_estimates/
[I have only read Elizabeth’s comment that I’m responding to here (so far); apologies if it would have been less confusing for me to read the entire thread before responding.]
I have always capitalized both EA and Rationality, and have never thought about it before. The first justification for capitalizing R that comes to mind is all the intentionality/intelligence that I perceive was invested into the proto-“AI Safety” community under EY’s (and others’) leadership. Isn’t it fair to describe the “Rationalist/Rationality” community as the branch of AI Safety/X-risk that is downstream of MIRI, LW, the Sequences, 🪄HPMOR, etc?