The objections section feels pretty strawmanny here.
1. It empowers your enemies
True. But it empowers your friends as well. Truth is an asymmetric weapon; if what you are working on is good, you will have more friends than enemies.
In the linked essay, truth is called "assymetric weapon" in the context of a logical debate. Unfortunately, the vast majority of the world doesn't form opinions based on logical debate. Storytelling and rhetoric prevail, especially in the short term.
For storytelling, truth (or rather data) is also asymmetric but in the exact opposite way. The more you share (especially of "raw" internal comms), the easier it is to cherrypick parts that make you look bad.
Debunking misleading claims takes significantly more effort than making those claims in the first place. Just take a look at the rise of alt-right or anti-vaccine movement.
The objections you stated as 2, 4 and 5 are logically flowing from this:
(Objection 1) Your enemies can use your transparency against you -> (4) this can damage the reputation of the org and anyone who works with it -> a mix of two outcomes happens:
(2) The org accepts the risk, this can sometimes result in materially bad outcomes (e.g. withdrawn funding or another org refusing to collaborate due to perceived hazard)
(5) The org (and the people in it) tries to mitigate the risk by investing more time to craft communication in a way that is less prone to be misused
I think that especially the last point is severely underestimated by calls for radical transparency. To be efficient, internal comms hugely rely on shared context, mutual good faith and short loop for asking clarifying questions. External comms have none of those benefits so crafting them takes significantly more effort - one needs to explain the context, pre-empt questions and account for the possible bad faith interpretations.
Some thoughts from a product manager working in tech:
AI increasing productivity allows startups to do more with less. I wouldn't necessarily expect it to significantly accelerate growth in general as coding speed is rarely the bottleneck for scaling products. You still need to grow your team, reach your target audience, find product-market fit, build your brand, convince customers to change habits, etc. This all takes time and effort. Coding is a surprisingly small part of it in most cases, especially for startups that aren't lacking funds for hiring engineers (which I'd expect to encompass ~all of the YC-backed ones).
None of your metrics is actually measuring the rate of growth. You are measuring the absolute size (valuation) of a company after a certain amount of time passes. It's entirely possible that the startups are indeed growing faster but are smaller than before, sold off earlier or taking less investment (making the post-YC valuation less accurate). A few effects that would incentivize this: (1) AI enables doing the same work with a smaller team, decreasing the need for external investment and therefore founder dilution, lowering founder incentives for high valuation (for a founder 30% share in a $50M company is actually better than a 10% share in $150M one, as they maintain more steering power) (2) AI makes copying a product easier, increasing the bargaining power of larger incumbents offering acquisition (3) Advances in general-purpose AI are reshaping markets and societies faster than any technological change before, increasing the risk of a startup becoming irrelevant despite early success.
AI capable of significantly accelerating coding in non-trivial use cases has only been available since mid 2025 - early 2026, depending on who you ask. The benefits of AI for coding that we're discussing today would not be visible in your data yet anyway.
I think this is looking too narrowly at the talent programmes' potential impact. To my understanding their overarching goal is getting more high-quality talent working on AI Safety faster. Counterfactually getting people hired is not the only action that can contribute to this. Off the top of my head I can think of:
Getting people hired counterfactually faster (as called out by @Ryan Kidd below)
Pre-filtering applicants for orgs with limited recruiting manpower (someone getting into a competitive programme is a decent signal, in the same way as a prestigious university or work at FAANG)
Reaching out to talent outside of the ecosystem who wouldn't even consider AI Safety otherwise
Connecting people who then go on to build new organizations
Getting people hired into counterfactually higher-impact positions than they would otherwise go for (e.g. many people seem to default to technical research even while they have transferrable skills to work in a more senior role in ops or governance)
To add to that comparison: Google invests into actively helping the applicants get in. They:
Have a standardized recruitment process that applicants can prepare for ahead of time
Point at resources which can be used to prepare
Offer mock interviews
Offer a grade of sorts, even to failing candidates (e.g. "Your interviews would put you at L5 level but right now we are looking for L6+")
Offer detailed feedback
Actively reach out to past candidates
This is based on my experiences with Google in Europe, might not generalize to the US. At the same time I'd expect Europe to be more talent-constrained, i.e. closer to the realities of the AI Safety ecosystem.
maybe previously a charity would have sold 20% of their impact for $1 million, but now they sell it for $5 million because lots of impact investors are competing.
What you are describing here is demand for funding shifting due to increased supply. But that demand isn't infinitely elastic. Scaling an organization is much more complex than just throwing money at it (this recent article has a good overview of the problem). That's why I'm talking about a demand-constrained market.
charities would also be able to set their own terms, so they could choose between selling 20% for $5 million or a smaller amount for $2 million.
I don't understand that part. How would they pick the number? To my understanding the "impact credits" are internet points with no inherent value.
If I understand the 1/(t+5) schedule correctly, it means the price of capital can't move in response to supply. In the for-profits investment markets it does: more investors competing to invest = less equity for the same dollars, because capital that's easy to replace has less counterfactual value.
If AI money suddenly makes funding abundant while vetting capacity doesn't grow accordingly (which seems to be the most likely scenario right now), we end up precisely in this sort of demand-constrained market. I'm curious how you'd adjust to account for that.
This article was brought to my attention by a discussion with @Austinhere. My main critique is in line with @harfe, cross-posting here:
[In the proposed impact market model] An org's valuation is implied by how much money it raises later. This measures how well a funder anticipated other funders, not the actual impact.
At a surface level this appears congruent with how the for-profit world operates. Startup valuations are also set by the money raised in later rounds. However there's a crucial difference - the value of firms is ultimately grounded in the expectations of future output in a way that value of non-profits can't be. Equity prices are guesses about very concrete future cash flows from people buying things the firm offers. The accuracy of those guesses can be ultimately checked against something that isn't another investor's opinion. The output is uniform - a dollar earned on a grocery store chain is the same as a dollar earned on an AI company[1].
In contrast, EA orgs aim for "positive impact" which can be quantified only by a very subjective evaluation with multiple assumptions, not by any market mechanism. The output is not uniform. Even if we had perfect measurement (we don't) I might value 1 human QALY above 1000 shrimp QALYs while you do the opposite.
This results in money invested being a much worse proxy for outcomes in a non-profit world than in the for-profit one. If I invest $1 M in Uber instead of Intel, I signal to everyone that I believe Uber is more likely than Intel to be worth more in some time horizon. If I invest $1 M in AMF instead of AWF, the signal is not clear. I might consider AMF to be more efficient than AWF but I might instead assign a higher value to human lives or simply have funds which are earmarked for GHD.
At least to a good approximation. Individual investors may choose to price in externalities, e.g. treat a grocery store dollar as "better" than the same dollar coming from an oil or tobacco company.
I think we're talking past each other on one specific thing: input dollars and output dollars aren't the same dollars.
Firms compete for earned dollars. That's the output, and it's what credits the firm for the value it creates (by spending money, customers signal that they value what the firm provides above its monetary cost). Invested dollars are an input. They happen to share a unit with the output, but that doesn't mean that they measure the same thing.
I read the recent substack post on impact markets and I think this is where it runs into trouble. An org's valuation is implied by how much money it raises later. This measures how well a funder anticipated other funders, not the actual impact.
At a surface level this appears congruent with how the for-profit world operates. Startup valuations are also set by the money raised in later rounds. However there's a crucial difference - the value of firms is ultimately grounded in the expectations of future output in a way that value of non-profits can't be. Equity prices are guesses about very concrete future cash flows from people buying things the firm offers. The accuracy of those guesses can be ultimately checked against something that isn't another investor's opinion. The output is uniform - a dollar earned on a grocery store chain is the same as a dollar earned on an AI company[1].
In contrast, EA orgs aim for "positive impact" which can be quantified only by a very subjective evaluation with multiple assumptions, not by any market mechanism. The output is not uniform. Even if we had perfect measurement (we don't) I might value 1 human QALY above 1000 shrimp QALYs while you do the opposite.
This results in money invested being a much worse proxy for outcomes in a non-profit world than in the for-profit one. If I invest $1 M in Uber instead of Intel, I signal to everyone that I believe Uber is more likely than Intel to be worth more in some time horizon. If I invest $1 M in AMF instead of AWF, the signal is not clear. I might consider AMF to be more efficient than AWF but I might instead assign a higher value to human lives or simply have funds which are earmarked for GHD.
Coming back to your original post - it considers Money, Taste and Dealflow as separate components, but in our discussion you seem to propose to base the attribution primarily on Money. VC doesn't do that. Limited Partners supply the capital and get returns proportional to capital. General Partners supply taste and dealflow and get carry. The carry exists precisely because "whose money was it" and "who found the deal" are understood to be different things. A regrantor is a GP, not an LP.
So I stand by my point that it's more relevant to look at how attribution works inside firms (which also cannot rely on markets here) than to default to the amount of money invested, just because it happens to come in a familiar unit.
At least to a good approximation. Individual investors may choose to price in externalities, e.g. treat a grocery store dollar as "better" than the same dollar coming from an oil or tobacco company.
Build low-commitment on-ramps: paid advisory engagements, board seats, fractional/interim senior roles, and short fellowships that let a mid-career professional contribute without a full career leap.
Is there evidence that "fractional/interim senior roles and short fellowships" actually help bring mid-career professionals on board? My intuition as a mid-career professional is the opposite - quitting my job to take on a time-bound opportunity feels significantly riskier than taking a permanent position. Double so if I'm aware that the ecosystem I'm trying to join has a strong preference for hiring from within.
nonprofit grantmakers should approximate their own assessment of impact as a function of "what fraction of the best nonprofits did I fund? For what size? How early?" - aka the same considerations that a for-profit vc considers for return on investment.
This approach only makes sense if you consider your available funding as a VC/grantmaker to be ~unlimited. In the VC world this might work for a16z or YC. Your regular VC fund/angel investor doesn't ultimately care what fraction of good deals they pass on or how early they joined on a deal but what literal ROI they get: net profit divided by money invested.
The objections section feels pretty strawmanny here.
In the linked essay, truth is called "assymetric weapon" in the context of a logical debate. Unfortunately, the vast majority of the world doesn't form opinions based on logical debate. Storytelling and rhetoric prevail, especially in the short term.
For storytelling, truth (or rather data) is also asymmetric but in the exact opposite way. The more you share (especially of "raw" internal comms), the easier it is to cherrypick parts that make you look bad.
Debunking misleading claims takes significantly more effort than making those claims in the first place. Just take a look at the rise of alt-right or anti-vaccine movement.
The objections you stated as 2, 4 and 5 are logically flowing from this:
(Objection 1) Your enemies can use your transparency against you ->
(4) this can damage the reputation of the org and anyone who works with it -> a mix of two outcomes happens:
I think that especially the last point is severely underestimated by calls for radical transparency. To be efficient, internal comms hugely rely on shared context, mutual good faith and short loop for asking clarifying questions. External comms have none of those benefits so crafting them takes significantly more effort - one needs to explain the context, pre-empt questions and account for the possible bad faith interpretations.
Some thoughts from a product manager working in tech:
A few effects that would incentivize this:
(1) AI enables doing the same work with a smaller team, decreasing the need for external investment and therefore founder dilution, lowering founder incentives for high valuation (for a founder 30% share in a $50M company is actually better than a 10% share in $150M one, as they maintain more steering power)
(2) AI makes copying a product easier, increasing the bargaining power of larger incumbents offering acquisition
(3) Advances in general-purpose AI are reshaping markets and societies faster than any technological change before, increasing the risk of a startup becoming irrelevant despite early success.
I think this is looking too narrowly at the talent programmes' potential impact. To my understanding their overarching goal is getting more high-quality talent working on AI Safety faster. Counterfactually getting people hired is not the only action that can contribute to this. Off the top of my head I can think of:
To add to that comparison: Google invests into actively helping the applicants get in. They:
This is based on my experiences with Google in Europe, might not generalize to the US. At the same time I'd expect Europe to be more talent-constrained, i.e. closer to the realities of the AI Safety ecosystem.
What you are describing here is demand for funding shifting due to increased supply. But that demand isn't infinitely elastic. Scaling an organization is much more complex than just throwing money at it (this recent article has a good overview of the problem). That's why I'm talking about a demand-constrained market.
I don't understand that part. How would they pick the number? To my understanding the "impact credits" are internet points with no inherent value.
If I understand the 1/(t+5) schedule correctly, it means the price of capital can't move in response to supply. In the for-profits investment markets it does: more investors competing to invest = less equity for the same dollars, because capital that's easy to replace has less counterfactual value.
If AI money suddenly makes funding abundant while vetting capacity doesn't grow accordingly (which seems to be the most likely scenario right now), we end up precisely in this sort of demand-constrained market. I'm curious how you'd adjust to account for that.
This article was brought to my attention by a discussion with @Austin here. My main critique is in line with @harfe, cross-posting here:
At least to a good approximation. Individual investors may choose to price in externalities, e.g. treat a grocery store dollar as "better" than the same dollar coming from an oil or tobacco company.
I think we're talking past each other on one specific thing: input dollars and output dollars aren't the same dollars.
Firms compete for earned dollars. That's the output, and it's what credits the firm for the value it creates (by spending money, customers signal that they value what the firm provides above its monetary cost). Invested dollars are an input. They happen to share a unit with the output, but that doesn't mean that they measure the same thing.
I read the recent substack post on impact markets and I think this is where it runs into trouble. An org's valuation is implied by how much money it raises later. This measures how well a funder anticipated other funders, not the actual impact.
At a surface level this appears congruent with how the for-profit world operates. Startup valuations are also set by the money raised in later rounds. However there's a crucial difference - the value of firms is ultimately grounded in the expectations of future output in a way that value of non-profits can't be. Equity prices are guesses about very concrete future cash flows from people buying things the firm offers. The accuracy of those guesses can be ultimately checked against something that isn't another investor's opinion. The output is uniform - a dollar earned on a grocery store chain is the same as a dollar earned on an AI company[1].
In contrast, EA orgs aim for "positive impact" which can be quantified only by a very subjective evaluation with multiple assumptions, not by any market mechanism. The output is not uniform. Even if we had perfect measurement (we don't) I might value 1 human QALY above 1000 shrimp QALYs while you do the opposite.
This results in money invested being a much worse proxy for outcomes in a non-profit world than in the for-profit one. If I invest $1 M in Uber instead of Intel, I signal to everyone that I believe Uber is more likely than Intel to be worth more in some time horizon. If I invest $1 M in AMF instead of AWF, the signal is not clear. I might consider AMF to be more efficient than AWF but I might instead assign a higher value to human lives or simply have funds which are earmarked for GHD.
Coming back to your original post - it considers Money, Taste and Dealflow as separate components, but in our discussion you seem to propose to base the attribution primarily on Money. VC doesn't do that. Limited Partners supply the capital and get returns proportional to capital. General Partners supply taste and dealflow and get carry. The carry exists precisely because "whose money was it" and "who found the deal" are understood to be different things. A regrantor is a GP, not an LP.
So I stand by my point that it's more relevant to look at how attribution works inside firms (which also cannot rely on markets here) than to default to the amount of money invested, just because it happens to come in a familiar unit.
At least to a good approximation. Individual investors may choose to price in externalities, e.g. treat a grocery store dollar as "better" than the same dollar coming from an oil or tobacco company.
Is there evidence that "fractional/interim senior roles and short fellowships" actually help bring mid-career professionals on board? My intuition as a mid-career professional is the opposite - quitting my job to take on a time-bound opportunity feels significantly riskier than taking a permanent position. Double so if I'm aware that the ecosystem I'm trying to join has a strong preference for hiring from within.
As a bit of a tangent:
This approach only makes sense if you consider your available funding as a VC/grantmaker to be ~unlimited. In the VC world this might work for a16z or YC. Your regular VC fund/angel investor doesn't ultimately care what fraction of good deals they pass on or how early they joined on a deal but what literal ROI they get: net profit divided by money invested.