Thanks for the post - I really appreciated hearing a perspective in favour of this overlap!
I just wanted to elaborate a bit on how "conflict of interest", "missing opportunities" and "all eggs in one basket" can come together to make the experience especially harmful.
In particular, if you face interpersonal issues (or worse e.g. harassment etc.) from someone who is in the same social graph (professionally and socially) as you, you're put in a very tricky spot. As you acknowledge, it is in fact how it works that if you aren't plugged into the social community, you will miss out on opportunities. It is also the case that many of your friends will be in the same social graph.
Thus you may be put in a position where you have to trade-off having less impact and not attending social events which all of your friends attend, versus your personal happiness and safety. While I think this possibility is very much implicit in your comments, I thought it might be useful to highlight it, because I think in practice this pressure to associate socially is very strong and therefore can be very damaging in the especially egregious cases of interpersonal issues within the community.
I find it curious you fail to engage with any of the other responses, and consider your time better spent on looking for dirt on people's Facebooks and Linkedins, especially since you claim:
This is a sincere question. Not intended to cause controversy.
That notwithstanding, I was not attempting to be patronising. I simply expected people (especially SPARC alum) who make strong claims on the forum to have a strong evidence base backing those claims up - for example, you question the validity of internal Atlas polling on the basis of respondents lying and bad survey methodology.
I'm assuming based off of your reply that your beliefs are influenced more by your priors than by having well-evidenced research or data on this, which is fine, but in that case, I think it would have been useful if you had flagged that level of epistemic confidence or rephrased your assertive claims in the original post!
On the object level, none of these seem to be great priors. There's no evidence for what non-trivial's scholarships have attracted, it's not clear that any of the projects you listed in the twitter thread about Stripe Atlas are the sort of thing Atlas would be trying to attract and in the case of Emergent Ventures, the "s" is doing a lot of work for you, since most grants are significantly higher than $1,000 and closer to the order of $10,000.
I was pretty skeptical of Atlas when I first heard about it, and I'm still concerned about overspending, bad optics, attracting grifters etc. That notwithstanding, this post doesn't even attempt to engage with what a programme with Atlas's goals should look like, and instead just reads as a long list of speculative gesturing.
Why do high schoolers need $50k scholarships?
In part, this makes sense as part of identifying talent. It's not obvious what the elasticity of grant money to attracted talent is, but it would be surprising if the elasticity is zero. As you note yourself, however, the money is also there for "academic expenses". Presumably identifying top talent is only an instrumental goal, and it is just as important to help said talent utilise their skills once you find them?
For many teenagers, they are likely financially constrained and this sort of funding can let them attend academic conferences or pay for textbooks etc. While it is true that they could in principle apply to EAIF or LTFF for these sorts of expenses, the website suggests that Atlas evaluates "the use of scholarship funds ... on a case-by-case basis", so this wouldn't be carte blanche for Atlas fellows to spend, and instead merely represents the diversification of funders within the community, which I see as a good thing.
In any case, my best guess is that the discretionary nature of this spending means that topline number of $50k is an upper bound which does not reflect the median Atlas fellow.
Of course, this leads to the next consideration:
Why was a $50k scholarship offered if a $25k scholarship would attract, say, 80-90% of the same applicants?
Would it?
I personally would be in favour of a smaller amount of money (somewhere in the range of $10k to $30k is my best guess), but I don't have good data on this. My impression is that many similar grant schemes offer comparable amounts of money (see: 776 Fellowship, O’Shaughnessy Fellowship, Thiel Fellowship, Century Fellowship) and those which offer less money often do so because they are aiming to provide funding for a much more limited time interval.
I suspect that a $5k unconditional grant that they can spend on whatever would attract just as many quality applications.
Given you seem to know a lot about optimal models of funding, I'd love to see the data you have which makes you believe this!
Many Atlas Fellows do not require the scholarship ... many fellows seem to be identified through traditional methods
In a similar vein to my previous comment, I'd be curious where you're getting your data from, and would love if you could publicise the survey you must undoubtedly have done to come to these conclusions, since you are not yourself at Atlas fellow!
I'd also point out that if the average Atlas fellow does not require the money, then the money needed to attract them is higher, so this point and the previous point seem in tension.
Atlas is planning on running a school called Atlas Academy Beta, as outlined in this document. Why does this make sense?
The document you link seems to address this, when it says that "we suspect that for many Atlas Fellows, the things holding them back from whatever ambitious and genuinely impactful projects they’re most interested in is a lack of deep, specialized technical knowledge—and the strict requirement to spend 8 hours in a classroom every day".
I'm not sure what the goals of the organisers of Atlas are, but judging by the funding structure and pop-up programmes like this, my guess is that they're not aiming for just talent search but also helping nurture the talent they find, which explains why they may diverge with programmes you've been involved with like SPARC.
As a final note, I find it a bit disconcerting that this post ends by speculating about the relationship status and preferences of people in the community, and I generally think that organising events or making grants does not represent consent in having people gossip about your personal life.
"having strong or intimate connections with employees of Open Philanthropy greatly enhances the chances of having funding, and it seems almost necessary"
Is this a well-identified phenomenon (in the causal inference sense) ?
Consider the following directed acyclic graph:
Connected with OpenPhil employees ----------> Gets funding from OpenPhil
^ ^ | | | | | | | | +------ Works on alignment ------+
One explanation for this correlation you identify is that being connected with OpenPhil employees leads to people getting funding from OpenPhil (as demonstrated by the horizontal arrow). However, another explanation is that working on alignment causes one to connect with others who are interested in the problem of AI alignment, as well as getting funding from a philanthropic organisation which funds work on AI alignment (as demonstrated by the vertical arrows).
These two explanations are observationally equivalent, in the absence of exogenous variation with respect to how connected one is to OpenPhil employees. Since claiming that it is "almost necessary" to have "strong or intimate connections with employees of Open Philanthropy" to get funding implies wrongdoing from OpenPhil, I'd be interested in evidence which would demonstrate this!
This argument proves too much. The same could be said of "go and donate your money, this (list of charities we think are most effective) is the way to do it".
My takeaway was that messages which could be spread include: "we should worry about conflict between misaligned AI and all humans", "AIs could behave deceptively, so evidence of safety might be misleading, "AI projects should establish and demonstrate safety (and potentially comply with safety standards) before deploying powerful systems", "alignment research is prosocial and great" and "we’re not ready for this". (I excluded "it might be important for companies and other institutions to act in unusual ways", because I agree this doesn't seem like a straightforward message to spread).
The answer is probably (a).
"Disproportionate" seems like it boils down to an object-level disagreement about relative cause prioritisation between AI safety and other causes.
Thanks for the post - I really appreciated hearing a perspective in favour of this overlap!
I just wanted to elaborate a bit on how "conflict of interest", "missing opportunities" and "all eggs in one basket" can come together to make the experience especially harmful.
In particular, if you face interpersonal issues (or worse e.g. harassment etc.) from someone who is in the same social graph (professionally and socially) as you, you're put in a very tricky spot. As you acknowledge, it is in fact how it works that if you aren't plugged into the social community, you will miss out on opportunities. It is also the case that many of your friends will be in the same social graph.
Thus you may be put in a position where you have to trade-off having less impact and not attending social events which all of your friends attend, versus your personal happiness and safety. While I think this possibility is very much implicit in your comments, I thought it might be useful to highlight it, because I think in practice this pressure to associate socially is very strong and therefore can be very damaging in the especially egregious cases of interpersonal issues within the community.
Your tone is very patronising here. I suggest rewording this.
I find it curious you fail to engage with any of the other responses, and consider your time better spent on looking for dirt on people's Facebooks and Linkedins, especially since you claim:
That notwithstanding, I was not attempting to be patronising. I simply expected people (especially SPARC alum) who make strong claims on the forum to have a strong evidence base backing those claims up - for example, you question the validity of internal Atlas polling on the basis of respondents lying and bad survey methodology.
I'm assuming based off of your reply that your beliefs are influenced more by your priors than by having well-evidenced research or data on this, which is fine, but in that case, I think it would have been useful if you had flagged that level of epistemic confidence or rephrased your assertive claims in the original post!
On the object level, none of these seem to be great priors. There's no evidence for what non-trivial's scholarships have attracted, it's not clear that any of the projects you listed in the twitter thread about Stripe Atlas are the sort of thing Atlas would be trying to attract and in the case of Emergent Ventures, the "s" is doing a lot of work for you, since most grants are significantly higher than $1,000 and closer to the order of $10,000.
I was pretty skeptical of Atlas when I first heard about it, and I'm still concerned about overspending, bad optics, attracting grifters etc. That notwithstanding, this post doesn't even attempt to engage with what a programme with Atlas's goals should look like, and instead just reads as a long list of speculative gesturing.
In part, this makes sense as part of identifying talent. It's not obvious what the elasticity of grant money to attracted talent is, but it would be surprising if the elasticity is zero. As you note yourself, however, the money is also there for "academic expenses". Presumably identifying top talent is only an instrumental goal, and it is just as important to help said talent utilise their skills once you find them?
For many teenagers, they are likely financially constrained and this sort of funding can let them attend academic conferences or pay for textbooks etc. While it is true that they could in principle apply to EAIF or LTFF for these sorts of expenses, the website suggests that Atlas evaluates "the use of scholarship funds ... on a case-by-case basis", so this wouldn't be carte blanche for Atlas fellows to spend, and instead merely represents the diversification of funders within the community, which I see as a good thing.
In any case, my best guess is that the discretionary nature of this spending means that topline number of $50k is an upper bound which does not reflect the median Atlas fellow.
Of course, this leads to the next consideration:
Would it?
I personally would be in favour of a smaller amount of money (somewhere in the range of $10k to $30k is my best guess), but I don't have good data on this. My impression is that many similar grant schemes offer comparable amounts of money (see: 776 Fellowship, O’Shaughnessy Fellowship, Thiel Fellowship, Century Fellowship) and those which offer less money often do so because they are aiming to provide funding for a much more limited time interval.
Given you seem to know a lot about optimal models of funding, I'd love to see the data you have which makes you believe this!
In a similar vein to my previous comment, I'd be curious where you're getting your data from, and would love if you could publicise the survey you must undoubtedly have done to come to these conclusions, since you are not yourself at Atlas fellow!
I'd also point out that if the average Atlas fellow does not require the money, then the money needed to attract them is higher, so this point and the previous point seem in tension.
The document you link seems to address this, when it says that "we suspect that for many Atlas Fellows, the things holding them back from whatever ambitious and genuinely impactful projects they’re most interested in is a lack of deep, specialized technical knowledge—and the strict requirement to spend 8 hours in a classroom every day".
I'm not sure what the goals of the organisers of Atlas are, but judging by the funding structure and pop-up programmes like this, my guess is that they're not aiming for just talent search but also helping nurture the talent they find, which explains why they may diverge with programmes you've been involved with like SPARC.
As a final note, I find it a bit disconcerting that this post ends by speculating about the relationship status and preferences of people in the community, and I generally think that organising events or making grants does not represent consent in having people gossip about your personal life.
Is this a well-identified phenomenon (in the causal inference sense) ?
Consider the following directed acyclic graph:
Connected with OpenPhil employees ----------> Gets funding from OpenPhil
^ ^
| |
| |
| |
| |
+------ Works on alignment ------+
One explanation for this correlation you identify is that being connected with OpenPhil employees leads to people getting funding from OpenPhil (as demonstrated by the horizontal arrow). However, another explanation is that working on alignment causes one to connect with others who are interested in the problem of AI alignment, as well as getting funding from a philanthropic organisation which funds work on AI alignment (as demonstrated by the vertical arrows).
These two explanations are observationally equivalent, in the absence of exogenous variation with respect to how connected one is to OpenPhil employees. Since claiming that it is "almost necessary" to have "strong or intimate connections with employees of Open Philanthropy" to get funding implies wrongdoing from OpenPhil, I'd be interested in evidence which would demonstrate this!
These don't seem very compelling to me.