The most common type is various instances of "utilitarianism endorses doing this thing that clearly decreases utility, therefore utilitarianism is wrong." Hard to remember specifics because this was 6 to 10 years ago. I just remember being struck by how these supposed experts had such basic misunderstandings.

FWIW, I didn't major in ethics but I did take a few ethics classes, and I found that every professor I saw had basic, obvious misunderstandings of utilitarianism.

This is a very cool project!

Have you looked into Idris? It has at least some of the capabilities that you'd want in a CEA language.

I still haven't looked much at Pedant, but I'm inclined to favor a DSL on top of a pre-existing language rather than a new language that requires its own compiler, largely because the former will be much easier to maintain and should be more portable—you're offloading all the work of writing the compiler to someone else. A custom language will indeed have a much simpler compiler, but the problem is you have to write and maintain the compiler yourself.

Also, possibly of interest to you: About five years ago I wrote a CEA DSL on top of C++. It's definitely not optimal—it's hard to use and only supports certain types of probability distributions—but it works for limited purposes. (I originally wrote it in Excel using VBA, but it was intolerably slow so I rewrote it in C++.)

No comment on the specific arguments given, but I like the way this post is structured: a list of weak arguments, grouped into categories, each short enough that they're easy to read quickly.

When I talked to an AI safety grad student about this, he said that Top 4 CS programs are not funding constrained, but top 10-20 are somewhat.

I've never been a grad student, but I suspect that CS grad students are constrained in ways that EA donors could fairly easily fix. They might not be grant-funding-constrained, but they're probably make-enough-to-feel-financially-secure-constrained or grantwriting-time-constrained, and you could convert AI grad students into AI safety grad students by lifting these constraints for them.

To the extent that you think good operations can emerge out of replicable processes rather than singularly talented ops managers, do you think it would be useful to write a longer article about how RP does operations? (Or perhaps you've already written this and I missed it)

Isn't housing more relevant than groceries? A typical household spends about 3x as much on housing than on all consumption goods combined (IIRC). And that site says housing in Nassau is a lot cheaper than in SF.

Hey, in hindsight I realize that the paper + summarization don't make clear that this does depend on model assumptions/empirical points

FWIW this was clear to me, I was using "conclusions" to mean "conclusions, given the model assumptions", not "conclusions, which the authors definitely think are true".

Yes this is something worth considering. I did look at how much alpha the Cambridge Associates startup data had on top of US publicly-traded tech stocks vs. the US total market, and there wasn't much difference. EA money is in much more specific investments than just the tech sector, but that makes it harder to test the correlation.

I would not consider Stripe a startup for the purposes of this post.