Ozzie Gooen
Bio
I'm currently researching forecasting and epistemics as part of the Quantified Uncertainty Research Institute.
Posts 61
Comments696
Topic contributions1
I really don't like the trend of posts saying that "EA/EAs need to | should do X or Y".
EA is about cost-benefit analysis. The phrases need and should implies binaries/absolutes and having very high confidence.
I'm sure there are thousands of interventions/measures that would be positive-EV for EA to engage with. I don't want to see thousands of posts loudly declaring "EA MUST ENACT MEASURE X" and "EAs SHOULD ALL DO THING Y," in cases where these mostly seem like un-vetted interesting ideas.
In almost all cases I see the phrase, I think it would be much better replaced with things like;
"Doing X would be high-EV"
"X could be very good for EA"
"Y: Cost and Benefits" (With information in the post arguing the benefits are worth it)
"Benefits|Upsides of X" (If you think the upsides are particularly underrepresented)"
I think it's probably fine to use the word "need" either when it's paired with an outcome (EA needs to do more outreach to become more popular) or when the issue is fairly clearly existential (the US needs to ensure that nuclear risk is low). It's also fine to use should in the right context, but it's not a word to over-use.
Excited to see this!
Really sorry, but recently we released Squiggle 0.8, which added a few features, but took away some things (that we thought were kind of footguns) that you used. So the model is now broken, but can easily be fixed.
I fixed it here, with some changes to the plots.
- I replaced "#" comments with "//" comments.
2. Instead of Plot.fn(), it's now Plot.distFn() or Plot.numericFn(), depending on if its returning a number or distribution. Note that these are now more powerful - you can adjust the axes scales, including adding custom symlog scales. I added a symlog yScale.
Also, in the newer version, you can specify function ranges. Like,
expected_number_of_scandals = {|n: [1e-3, 30k]|(1 / 5000 to 1 / 500) * n}You have the code,
growth_benefits = {|n: range|n * log(n)}This is invalid where n=0, so I changed the range to start slightly above that. (You can also imagine other ways of doing that)
In the future, we really only plan to have breaking changes in main version numbers, and we'll watch them on Squiggle Hub. I didn't see that there were recent active Squiggle models. Sorry for the confusion, again!
Quick things:
1. The vast majority of the funding comes from a very few funders. In the cases of GiveWell/ACE, they are meant more for many smaller donors. Historically, funding was not the main bottleneck in longtermism ("we might as well fund everything") - that said, this is changing now, so it could be a good time to do something different.
2. My guess is that few strong people wanted to themselves make a new evaluator in this space. Many of the strong people who could evaluate longtermist orgs seem to prefer working at OP or, more likely, just working in the field directly. EA is kind of a "do-ocracy" - often the reason for "why didn't this happen?" is, "it would require some strong talent, and none really was excited about this."
3. Similar to (1), I haven't seen a large funding base be very interested in spending money on evaluation here.
4. Much of the longtermist field is fairly friendly with one another and I think there's a gap of candid evaluation.
5. Honestly, I think a lot of opinions around specific longtermist interventions seem very intuition-based and kind of arbitrary. Especially around AI, there seem to be a bunch of key considerations that many people disagree about - so it's tricky to have a strong set of agreements to do evaluation around.
6. Some people seem to think that basically nothing in AI Safety is very promising yet, so we're just trying a bunch of stuff and hoping that eventually strong research programs are evident.
I'd like to see more work here and really hope that things mature.
Some thoughts:
I think that you can think about "forecasting" as one of a collection of intellectual practices that the EA community is unusually focused on.
Other practices/norms include:
- "Scout Mindset"
- Bayesian Thinking
- Lots of care/interest about analytical philosophy
- A preferences for empirical data/arguments
- A mild dislike for many kinds of emotions arguments
- Rationalism / Rationalist Fiction
I think that a background variable here is that EA is close to an intellectual community of thinkers that use similar tools and practices. Thinkers like Steven Pinker, Matthew Yglesias, Robin Hanson, Bryan Caplan, and Andrew Gelman come to mind as people with somewhat similar styles and interests. Many of these people also showed unusual interest in "forecasting".
So some questions here would be:
1. How well does EA fit into some outer intellectual tribes, like I hinted at above?
2. What preferences/norms do these tribes have, and how justified are they?
3. Why aren't all of these preferences/norms more common?
I think that "forecasting" as we discuss it is often a set of norms like:
1. Making sure that forecasts are recorded and scored.
2. Making sure forecasters are incentivized to do well.
3. Making sure that the top forecasters are chosen for future rounds.
To do this formally requires a fair bit of overhead, so it doesn't make much sense for small organizations.
I think that larger organizations either know enough to vet and incentivize their existing analysts (getting many of the benefits of forecasters), or don't, in which case they won't be convinced to use forecasters. (I think that obvious explanations are some of the reason, but I do have questions here)
Society-wide, I think most people don't care about forecasters for similar reasons that most people don't care about Bayesian Thinking, Scott Alexander, or Andrew Gelman. I think these tools/people are clever, but others aren't convinced/knowledgeable of them.
Open to this as well, if there are specific individuals you think would do this well as a fulltime job! (Though, in this model I'm not quite sure what the role of the regrantor is -- just to scout out opportunities?)
I get the impression there's a lot of human work that could be done to improve the process.
- Communicate with potential+historic recipients. Get information from them (to relay to grantmakers), and also inform them about the grantmaker's preferences.
- Follow up with grantees to see how progress went / do simple evaluations.
- Do due diligence into several of the options.
- Maintain a list of expert advisors that could be leaned on in different situations.
- Do ongoing investigations and coordination with funded efforts. I think a lot of value comes from funding relationships that last 3-10+ years.
> I agree EA is pretty strange, and think we could benefit from overhauling the ecosystem to be more like the tech scene, which seems to be much better than EA at executing at their objectives.
This is a long discussion. To me, a lot of the reason way startups work is that the tail outcomes are really important - important enough to justify lots of effective payment for talent early on. But this only makes sense under the expectations of lots of growth, which is only doable with what eventually need to be large organizations.
> I agree with you on the importance of organizational reform, though it's extra unclear if that kind of thing is addressable by a regranting program (vs someone enacting change from within OpenPhil). Perhaps we'll ourselves address this if/when Manifund itself represents a significant chunk of EA funding, but that seems like a version 2 problem while we're currently at v0.
My guess is that this depends a lot on finding the right grantmaker, and it's very possible that's just not realistic soon. I agree this can wait, there's a lot of stuff to do, just giving ideas to be thinking about for later.
> Would love to do a podcast/debate on this sometime!
That sounds fun! Maybe we could just try to record a conversation next time we might see each other in person.
Thanks for the replies! Quickly,
> but if other donors are looking for a more hands-on role, we'd be open to that as well!
My guess is that some donors don't exactly want to be hands-on for specific grants, but do want to get updates and ask specific questions to the grantmaker. This could be a bit of a pain to the grantmaker, but would be a better experience for the funder. In some cases, this seems worth it (mainly if we want to get more funders).
> trying to catch up to what other funders have figured out
For what it's worth, at the LTFF (and I think other EA Funds), there's a voting system where people vote on scores, and proposals that achieve a certain average or higher get funded. I agree this is overkill for many small grants.
> and I suspect FF's may have had bad incentives + optics, so we opted to launch first and revisit this later.
I like that strategy. Later on though, I'd imagine that maybe regranters could request different perks. If there were a marketplace - funders choose regranters - then some regranter could request ~5%, and funders would take that into consideration. I suspect some granters don't need the money, but others might only be able to do it if there were pay.
Just want to flag that I'm really happy to see this. I think that the funding space could really use more labor/diversity now.
Some quick/obvious thoughts:
- Website is pretty great, nice work there. I'm jealous of the speed/performance, kudos.
- I imagine some of this information should eventually be private to donors. Like, the medical expenses one.
- I'd want to eventually see Slack/Discord channels for each regrantor and their donors, or some similar setup. I think that communication between some regranters and their donors could be really good.
- I imagine some regranters would eventually work in teams. From being both on LTFF and seeing the FTX regrantor program, I did kind of like the LTFF policy of vote averaging. Personally, I think I do grantmaking best when working on a team. I think that the "regrantor" could be a "team leader", in the sense that they could oversee people under them.
- As money amounts increase, I'd like to see regranters getting paid. It's tough work. I think we could really use more part-time / full-time work here.
- I think if I were in charge of something like this, I'd have a back-office of coordinated investigations for everyone. Like, one full-time person who just gathers information about teams/people, and relays that to regranters.
- As I wrote about here, I'm generally a lot more enthusiastic about supporting sizeable organizations than tiny ones. I'd hope that this could be a good project to fund projects within sizeable organizations.
- I want to see more attention on reforming/improving the core aspects/community/bureaucracy of EA. These grantmakers seem very AI safety focused.
- Ideally it could be possible to have ratings/reviewers of how the regranters are to work with. Some grantmakers can be far more successful than others at delivering value to grantees and not being a pain to work with.
- I probably said this before, but I'm not very excited by Impact Certificates. More "traditional" grantmaking seems much better.
- One obvious failure mode is that regranters might not actually spend much of their money. It might be difficult to get good groups to apply. This is not easy work.
Good luck!
your counterargument here does not properly address the points from Middle Manager Hell / the Immoral Mazes sequences
I didn't mean for it to. I was just pointing at the general dislike for Middle Management.
More constructively, one difference between current EA structure and large orgs is that small EA orgs are not married to a single funder.
I think one awkward thing now is that many of them are sort of married to a single funder. The EA funding community can be really narrow right now - just a few funders, and with similar/overlapping opinions and personnel.
All that said, perhaps we can get the best of the both worlds by using larger orgs for some things but not all?
I think we can do better about getting better sets of trade-offs. There's also "large orgs, with lots of team freedom" as a medium. I also listed some strategies in the "Try to Mimic Organizational Strengths" section. I'd be happy to see this area worked on further!
I recently noticed it here:
https://forum.effectivealtruism.org/posts/WJGsb3yyNprAsDNBd/ea-orgs-need-to-tabletop-more
Looking back, it seems like there weren't many more very recently. Historically, there have been some.
EA needs consultancies
EA needs to understand its “failures” better
EA needs more humor
EA needs Life-Veterans and "Less Smart" people
EA needs outsiders with a greater diversity of skills
EA needs a hiring agency and Nonlinear will fund you to start one
EA needs a cause prioritization journal
Why EA needs to be more conservative
Looking above, many of those seem like "nice to haves". The word "need" seems over-the-top to me.