I'm interested in hearing from those who provided downvotes. I could imagine a bunch of reasons why one might have done so (there were a lot of points included here).  

Minor points, from your comment:

Yes we can withold funding, after the fact, and at great community reputational cost. But I can't e.g. get SBF to not do podcasts nor stop the EA (or two?) that seem to have joined DOGE and started laying waste to USAID.

I believe most EAs would agree these examples should never have been in OP's proverbial sphere of responsibility.

There are other examples we could discuss regarding OP's role (as makes sense, no organization is perfect), but that might distract from the main topic: clarity on the OP-EA relationship and the mutual expectations between parties.

On Bsky, they blame EAs for the whole endeavor

It seems obvious that such Bsky threads contain significant inaccuracies. The question is how much weight to give such criticisms.

My impression is that many EAs wouldn't consider these threads important enough to drive major decisions like funding allocations. However, the fact you mention it suggests it's significant to you, which I respect.

About the OP-EA relationship - if factors like "avoiding criticism from certain groups" are important for OP's decisions, saying so clearly is the kind of thing that seems useful. I don't want to get into arguments about if it should^[1], the first thing is to just understand that that's where a line is. 

1. ^{^}

   More specifically, I think these discussions could be useful - but I'm afraid they will get in the way of the discussions of how OP will act, which I think is more important. 

Thanks for the response here! I was not expecting that. 

This is a topic that can become frustratingly combative if not handled gracefully, especially in public forums. To clarify, my main point isn't disagreement with OP's position, but rather I was trying to help build clarity on the OP-EA relationship.

Some points:
1. The relationship between the "EA Community" and OP is both important (given the resources involved) and complex^[1] . 
2. In such relationships, there are often unspoken expectations between parties. Clarity might be awkward initially but leads to better understanding and coordination long-term.
3. I understand you're uncomfortable with OP being considered responsible for much of EA or accountable to EA. This aligns with the hypotheses in my original comment. I'm not sure we're disagreeing on anything here.
4. I appreciate your comments, though I think many people might reasonably still find the situation confusing. This issue is critical to many people's long-term plans. The links you shared are helpful but leave some uncertainty - I'll review them more carefully.
5. At this point, we might be more bottlenecked by EAs analyzing the situation than by additional writing from OP (though both are useful). EAs likely need to better recognize the limitations of the OP-EA relationship and consider what that means for the community.
6. When I asked for clarification, I imagined that EA community members working at the OP-EA intersection would be well positioned to provide insight. One challenge is that many people feel uncomfortable discussing this relationship openly due to the power imbalance.^[2]. As well as the funding issue (OP funds EA), there's also the fact that OP has better ways of privately communicating^[3]. (This is also one issue why I'm unusually careful and long with my words with these discussions, sorry if it comes across as harder to read.) That said, comment interactions and assurances from the OP do help build trust.

1. ^{^}

   there's a fair bit of nuance involved - I'm sure that you have noticed confusion on the side of EAs at least

2. ^{^}

   i.e. For example, say an EA community member writes something that upsets someone at OP. Then that person holds a silent grudge, decides they don't like that person, then doesn't fund them later. This is very human, and there's a clear information asymmetry. The EA community member would never know if this happens, so it would make sense for them to be extra cautious. 

3. ^{^}

   People at OP can confidentially discuss with each other how to best handle their side of the OP-EA relationship. But in comparison, EA community members mainly have the public EA Forum, so there's an inherent disadvantage. 

The loss of active donors seems like a big deal to me. I'm really curious what's happening there.

1. Some donors probably just changed targets, for example, to Manifund or other new funds
2. I'm sure some got cynical or disenchanted by the movement or something
3. Maybe some got hurt financially, from FTX / a crypto bust?

Someone pointed me to this blog post by the EA Infrastructure Fund, which I missed.

https://forum.effectivealtruism.org/posts/PN4FEGt3fjHv8diGo/eaif-isn-t-currently-funding-constrained

> EAIF currently has $3.3M in available funds. So far over 2024 EAIF has made grants worth $1.1M, and I expect this to be around $1.4M by the end of 2024. 

So it seems like EAIF at least is in a decent position. 

I'm not sure how this coincides with the above numbers though. Just eyeballing it - maybe, while donations in have been low, donations out have been lower, so there's been a surplus of funding. 

Do people know what's going on with the EA Funds financial position? Some of the graphs look a bit worrying. But I've also noticed some (what seem like) inconsistencies, so I suspect that some key data is missing or something. 

https://funds.effectivealtruism.org/stats/donations

The following graph seems incorrect. It seems like it just wasn't updated since Dec 2023. And I'm curious if there really was the ~$2M jump in the LTFF that month. 

Neat!

Quick point: The "Template Here" link doesn't have the right access.  

Occasionally I come across people who assume that "AI + judgemental forecasting" is a well-studies and funded area.

While I think this area is really high-potential (especially if done correctly), I think it's quite small right now.

I know of two new startups in the space (one is FutureSearch, I heard of another on Twitter). Neither has a very serious public product out yet, both probably have fewer than 10 FTEs. Based on the rates of startup success, I'd guess both will fail or pivot.

Metaculus has done some public challenges for AI forecasting, but these are pretty small. A $~30k prize.

There are occasional research papers that come out in the area. But these typically don't come with tools that the public can really use.

Overall, while I think this field is really neat, I think there are fairly few people actively working on it now, and I'd expect correspondingly limited results.

This reminds me of another related tension I've noticed. I think that OP really tries to not take much responsibility for EA organizations, and I believe that this has led to something of a vacuum of leadership. 

I think that OP has functionally has great power over EA. 

In many professional situations, power comes with corresponding duties and responsibilities.

CEOs have a lot of authority, but they are also expected to be agentic, to keep on the lookout for threats, to be in charge of strategy, to provide guidance, and to make sure many other things are carried out.

The President clearly has a lot of powers, and that goes hand-in-hand with great expectations and duties.

There's a version of EA funding where the top funders take on both leadership and corresponding responsibilities. These people ultimately have the most power, so arguably they're best positioned to take on leadership duties and responsibilities.

But I think nonprofit funders often try not to take much in terms of responsibilities, and I don't think OP is an exception. I'd also flag that I think EA Funds and SFF are in a similar boat, though these are smaller. 

My impression is that OP explicitly tries not to claim any responsibility for the EA ecosystem / environment, and correspondingly argues it's not particularly accountable to EA community members. Their role as I understand it is often meant to be narrow. This varies by OP team, but I think is true for the "GCR Capacity Building" team, which is closest to many "EA" orgs. I think this team mainly thinks of itself as a group responsible for making good decisions on a bunch of specific applications that hits their desk.

Again, this is a far narrower mandate than any conventional CEO would have. 

If we had a "CEO or President" that were both responsible for and accountable to these communities, I'd expect things like:
1. A great deal of communication with these communities.
2. Clear and open leadership structures and roles.
3. A good deal of high-level strategizing.
4. Agentic behavior, like taking significant action to "make sure specific key projects happen." 
5. When there are failures, acknowledgement of said failures, as well as plans to fix or change.

I think we basically don't have this, and none of the funders would claim to be this.

So here's a question: "Is there anyone in the EA community who's responsible for these sorts of things?"

I think the first answer I'd give is "no." The second answer is something like, "Well, CEA is sort of responsible for some parts of this. But CEA really reports to OP given their funding. CEA has very limited power of its own. And CEA has repeatedly try to express limits in its power, plus its gone through lots of management transitions."

In a well-run bureaucracy, I imagine that key duties would be clearly delegated to specific people or groups, and that groups would have the corresponding powers necessary to actually do a good job at them. You want key duties to be delegated to agents with the power to carry them out.

The ecosystem of EA organizations is not a well-organized bureaucracy. But that doesn't mean there aren't a lot of important duties to be performed. In my opinion, the fact that EA represents a highly-fragmented set of small organizations was functionally a decision by the funders (at least, they had a great deal of influence on this), so I'd hope that they would have thoughts on how to make sure the key duties get done somehow. 

This might seem pretty abstract, so I'll try coming up with some more specific examples:
1. Say a tiny and poorly-resourced org gets funded. They put together a board of their friends (the only people available), then proceed to significantly emotionally abuse their staff. Who is ultimately responsible here? I'd expect the founders would not at all want to take responsibility for this. 
2. Before the FTX Future Fund blew up, I assumed that EA leaders had vetted it. Later I find out that OP purposefully tried to keep its distance and not get involved (in this case meaning that they didn't investigate or warn anyone), in part because they didn't see it as their responsibility, and claimed that because FTX Future Fund was a "competitor", it wasn't right for them to get involved. From what I can tell now, it was no one's responsibility to vet the FTX Future Fund team or FTX organization. You might have assumed CEA, but CEA was funded by FTX and previously even had SBF as a board member - they were clearly not powerful and independent enough for this.
3. There are many people in the EA scene who invest large amounts of time and resources preparing for careers that only exist under the OP umbrella. Many or all of their future jobs will be under this umbrella. At the same time, it's easy to imagine that they have almost no idea what the power structures at the top of this umbrella are like. This umbrella could change leadership or direction at any time, with very little warning. 
4. There were multiple "EAs" on the board of OpenAI during that board member spat. That event seemed like a mess, and it negatively influenced a bunch of other EA organizations. Was that anyone's responsibility? Can we have any assurances that community members will do a better job next time? (if there is a next time)
5. I'm not sure if many people at all, in positions of power, are spending much time thinking about long-term strategic issues for EA. It seems very easy for me to imagine large failures and opportunities we're missing out on. This also is true for the nonprofit EA AI Safety Landscape - many of the specific organizations are too small and spread out to be very agentic, especially in cases of dealing with diverse and private information. I've heard good things recently about Zach Robinson at CEA, but also would note that CEA has historically been highly focused on some long-running projects (EAG, the EA Forum, Community Health), with fairly limited strategic or agentic capacity, plus being heavily reliant on OP. 
6. Say OP decides to shut down the GCR Capacity Building team one day, and gives a 2-years notice. I'd expect this to be a major mess. Few people outside OP understand how the internals of OP decisions get made, so it's hard for other EA members to see this coming or gauge how likely it is. My guess is that they don't seem like they'd do this, but I have limited confidence. As such, it's hard for me to suggest that people make long-term plans (3+ years) in this area. 
7. We know that OP generally maximizes expected value. What happens when narrow EV optimization conflicts with honesty and other cooperative values? Would they represent the same choices that other EAs might want? I believe that FTX justified their bad actions using utilitarianism, for instance, and lots of businesses and nonprofits carry out highly Machiavellian and dishonest actions to advance their interests. Is it possible that EAs working under the OP umbrella are unknowingly supporting actions they might not condone? It's hard to know without much transparency and evaluation. 

On the plus side, I think OP and CEA have improved a fair bit on this sort of thing in the last few years. OP seems to be working to assure that grantees follow certain basic managerial criteria. New hires and operations have come in, which has seemed to have helped. 

I've previously discussed my thinking on the potential limitations we're getting from having small orgs here. Also, I remember that Oliver Habryka has repeatedly mentioned the lack of leadership around this scene - I think that this topic is one thing he was sort-of referring to. 

Ultimately, my guess is that OP has certain goals they want to achieve, and it's unlikely they or the other funders will want to take many of the responsibilities that I suggest here. 

Given that, I think it would be useful for people in the EA ecosystem to understand this and respond accordingly. I think that our funding situation really needs diversification, and I think that funders willing to be more agentic in crucial areas that are currently lacking could do a lot of good. I expect that when it comes to "senior leadership", there are some significant gains to be made, if the right people and resources can come together.  

I think this is neat! It's also lengthy, I like the write-up.

Some quick thoughts:
1. I'd be curious if the source code or specific prompts that pgodzinai used are publicly. It seems like it took the author less than 40 hours, so maybe they could be paid for this, worst case. 
2. I find it interesting that the participants included commercial entities, academic researchers, etc. I'm curious if this means that there's a budding industry of AI forecasting tools.
3. It sounds like a lot of the bots are using similar techniques, and also seems like these techniques aren't too complicated. Here the fact that pgodzinai took fewer than 40 hours comes to mind. "The top bot (pgodzinai) spent between 15 and 40hr on his bot.". At very least, it seems like it should be doable to make a bot similar to pgodzinai and have it be an available open-source standard that others could begin experimenting with. I assume we want there to be some publicly available forecasting bots that are ideally close to SOTA (especially if this is fairly cheap, anyway). One thing this could do is act as a "baseline" for future forecasting experiments by others. 
4. I'm curious about techniques that could be used to do this for far more questions, like 100k questions. I imagine that there could be a bunch of narrower environments with limited question types, but in conditions where we could much more rapidly test different setups. 
5. I imagine it's a matter of time until someone can set up some RL environment that deeply optimizes a simple forecasting agent like this (though would do so in a limited setting).