I currently lead EA funds.
Before that, I worked on improving epistemics in the EA community at CEA (as a contractor), as a research assistant at the Global Priorities Institute, on community building, and Global Health Policy.
Unless explicitly stated otherwise, opinions are my own, not my employer’s.
You can give me positive and negative feedback here.
I agree with this take, but I think that this is primarily an agency and “permission to just do things” issue, rather than people not being good at prioritisation. It requires some amount of being willing to be wrong publicly and fail in embarrassing ways to actual go and explore under-explored things, and in general, I think that current EA institutions (including people on the EA forum) don’t celebrate people actually getting a bunch of stuff done and testing a bunch of hypothesis (otoh, shutting down your project is celebrated which is good imo—though ~only if you write a forum post about it).
I guess overall, I don’t think that people are as bottlenecked on how much prioritisation they are doing, but are pretty bottlenecked on “just doing things” and not staring into the void enough to realise that they should move on from their current project or pivot (even when people on the forum will get annoyed with them). In part, these considerations converge because it turns out that many projects are fairly bottlenecked by the quality of execution rather than by the ideas themselves, and actually trying out a project and iterating on it seems to reliably improve projects.
In general, my modal advice for EAs has shifted from “try to think hard about what is impactful” towards “consider just doing stuff now” or “suppose you have 10x more agency, what would you do? Maybe do that thing instead?”.
There’s an adjacent take I agree which is more like:
1. AI will likely create many high-stakes decisions and a confusing environment
2. The situation would be better if we could use AI to stay in-step with AI progress on our ability to figure stuff out
3. rather than waiting until the world is very confusing, maybe we should use AIs right now to do some kinds of intellectual writing, in ways we expect to improve as AIs improve (even if AI development isn’t optimising for intellectual writing).
I think this could look a bit like company with mostly AI workers that produces writing on a bunch of topics, or as a first step, heavily LM written (but still high-quality) substack.
I’d bet that current models with less than $ 100,000 of post-training enhancements achieve median human performance on this task.
Seems plausible the metaculus judges would agree, especially given that that comment is quite old.
I think the mainline plan looks more like use the best agents/model internally and release significantly less capable general agents/models, very capable but narrow agents/models, or AI generated products.
Yeah that’s fair. I’m a lot more bullish on getting AI systems that satisfy the linked question’s definition than my own one.
Did you look at the metaculus resolution criteria? They seem extremely weak to me, would be intersted to know which critiera you think o3 (or whatever the best OAI model is) is furthest away from.
AGI by 2028 is more likely than not
Most of my uncertainty is from potentially not understanding the criteria. They seem extremely weak to me:
* Able to reliably pass a Turing test of the type that would win the Loebner Silver Prize.
Able to score 90% or more on a robust version of the Winograd Schema Challenge, e.g. the “Winogrande” challenge or comparable data set for which human performance is at 90+%
Be able to score 75th percentile (as compared to the corresponding year’s human students; this was a score of 600 in 2016) on all the full mathematics section of a circa-2015-2020 standard SAT exam, using just images of the exam pages.
Be able to learn the classic Atari game “Montezuma’s revenge” (based on just visual inputs and standard controls) and explore all 24 rooms based on the equivalent of less than 100 hours of real-time play (see closely-related question.)
I wouldn’t be surprised if we’ve already passed this.
Bioweapons are an existential risk
Note that imo almost all the x-risk from bio routes through AI, and is better thought of as an AI-risk threat model.
You might believe future GPU hours are currently underpriced (e.g. maybe we’ll soon develop AI systems that can automate valuable scientific research). In such a scenario, GPU hours would become much more valuable, while standard compute credits (which iiuc are essentially just money designated for computing resources) would not increase in value. Buying the underlying asset directly might be a straightforward way to invest in GPU hours now before their value increases dramatically.
Maybe there are cleverer ways to bet on price of GPU hours dramatically increasing that are conceptually simpler than nvidia share prices increasing, idk.
I think that “moods” should be a property of the whole discourse, as opposed to specific posts. I find it a bit annoying when commenters say a specific post has a missing mood—most posts don’t aim to represent the whole discourse.
I think this kind of investigation would be valuable, but I’m not sure what concrete questions you’d imagine someone answering to figure this out.
Yeah, I think starting research nonprofits is an especially different skillset. Good point.
Fwiw, I am a fan of open-source public goods. I think more nonprofits should try to build products rather than papers (or at least be in that mindset). I am a little worried that doing research is just more accepted than producing products that people might use and that creates unhelpful barriers. Relatedly, I quite like the move towards essay websites more care goes towards thinking about the UX and trying to actually interact with users.
I mostly agree with this. I wasn’t really aiming to give a balanced take here on whether people should start for-profit instead of nonprofit—I just meant to list a few (imo) underrated by EAs features of for-profits.
I’m more confused about the poor personal fit point. I suspect that many EAs are also a bad fit (at least initially) for starting nonprofits, but the EA ecosystem makes it somewhat easier for nonprofits to survive (which imo is probably a good thing overall).
The version of your claim I most agree with is:
* An EAs comparative advantage is identifying the most overlooked and big deal moral considerations, and access to minimal viable nonprofit capital
* This selects for highly neglected things where you might be able to have an impact at a small scale, with not particularly high standards
* So even if EAs aren’t sufficiently competent to build high-growth companies, they can still have an outsized impact via founding nonprofits
Makes sense. Sorry I didn’t say this before but I really appreciate your comments on this post and the high effort modelling that you did when working out what to work on. I think it’s a great example to set for the community and shows how seriously you can take these decisions (if you want to).
I don’t quite understand why you’re comparing donated money from people earning to give to money consumed by a charity.
I think a better comparison could be to just convert everything into impact adjusted dollars imagining that you’re able to sell off your impact equity. In this scheme it’s clearer that taking more money from value aligned funders is bad, whilst taking money from non-aligned funders is roughly neutral and donating a bunch of money is very good. The charity has to essentially “pay back” in impact the value aligned funder to get itself out of a hole, whereas the for profit doesn’t need to worry about that (on impact grounds).
To be clear, I think many nonprofits do a lot of good and are well worth funding—I spend most of my time trying to fund them—but it is harder to work out net positive if you’re consuming a bunch of fungible and value aligned resources.
Ok, I did also write:
* Nonprofits have significantly weaker feedback mechanisms compared to for-profits
* Few people are going to complain that you provided bad service when it didn’t cost them anything.
* Most nonprofits are not very ambitious, despite having large moral ambitions.
* It’s challenging to find talented people willing to accept a substantial pay cut to work with you.
* For-profits are considerably more likely to create something that people actually want.
Which I think are all differentially more useful in short timelines than in long timelines worlds as you don’t have less time to mess around, and you instead need to very quickly work on a useful thing. If you disagree with this maybe we’re talking past other and I misunderstand your perspective.
Fwiw I’m sympathetic to nonprofits being better on net than for profit in short-timelines for reasons that aren’t discussed in this post e.g. greater freedom to focus narrowly on useful work, particularly in cases where there isn’t a viable business model.
I’m not sure why you wouldn’t care about standards if timelines are short. I feel like you should care more about people actually using your work, which might be more likely if you are forced to get product-market fit.
I think the way the question is framed is a bit leading. The optimal amount of democratically elected EA leadership is probably non-zero (idk lots of things are non-zero), but I struggled to come up with any examples of currently existing EA leadership positions that I felt should be elected.