Pity the grantmaker.

Today, they must have enough taste to decide the problem area, the most promising approach, and the team to do it. If they’re unlucky, they might have to deal with donors with quirky preferences, possess the patience to write many BOTECs and page-long grant investigations, and develop the on-the-ground information flow for potential opportunities (many of the most impactful of which are, understandably, not written about publicly on the “EA forum.”)

Today, there are a few different ways that grantmakers give money, that vary along a spectrum of “how prescriptive the grant-maker is.” Starting from least to most prescriptive:

Scrolling through a proposal aggregation site, like Manifund or grantmaking.ai, where people post publicly about their projects.
The humble open call for applications. Anyone can apply for anything that fits within the broad theme of the fund.
The “request for proposals“ (RFPs), where the grantmaker writes this little page about a specific sub-theme they’re looking to fund, and maybe includes some example projects.
Full-service “active grantmaking”, where grantmakers scope the project and make the initial hires.

Particularly on the more involved end: this is a lot to ask of our poor grantmakers, who, on one side, now need the creativity to come up with new organizations to found, and the operational networks to hire good leaders. As Sophie argues, we need greater specialization across the giving supply chain. I hope that Bluedot course in grantmaking fundamentals is pretty comprehensive.

Subscribe now

In this post, I’ll argue for a little proposal to alleviate these problems and more: conditional commitments. It’s very simple, a grantmaker (or a regrantor empowered to commit the fund’s money, not just recommend) has the ability to make a time-bound, high-confidence guarantee of specific funding for a project that passes certain conditions.

For example, a note that says “for the next 3 months, conditional on there being an executive director that passes my vibe check, I will fund this proposal for $3m.” (And yes, this was the structure of a real conditional commitment I helped run!)

The high-confidence, specific amount is the load-bearing feature that distinguishes this from a well-written RfP.

The general structure of the commitment is: “for [A time], conditional on [B conditions], I’ll fund [C proposal] for [$D].” (Unlike advance market commitments or prizes, which condition on outcomes, these condition on initial conditions -- the team exists, the institution says yes.)

*Hogarth, “Marriage A-la-Mode: The Marriage Settlement” (1743). Sometimes AI safety grantmaking has the drama of a family affair*

Compared to active grantmaking, conditional commitments solve the founder-funder chicken-egg problem.

Hiring for established organizations is already difficult, but getting someone to quit their job for an idea (particularly if it’s a stable, multi-decade one and they’re just about to get a promotion) or even “there’s a high likelihood we’d fund a proposal of this shape” is just so much worse than: “there’s $10m waiting if you -- yes, you, best person in the world to run this -- leave your job to do so.” This does make those cold outreach emails easier.
Similarly, there are many ideas that are great in the abstract, that many funders would be excited about, and for which the best person in the world is easily findable (or, at least, describable) -- but whose main risk is that no team exists to execute.
For institutions (eg. if you’re convincing a think tank to set up a new Center) or other funders, a commitment of this kind (”I have a donor with $Dm interested in starting something) opens doors and makes them immediately listen. Compare this with “I’m thinking of applying to the ‘CG Philanthropy’ RfP with this idea, speak to me?”

Conditional commitments also enable a new division of labor: between people with taste and people with time.

With regranting, at least naively designed, the people with the most taste don’t have the best information flow on potential opportunities. Conditional commitments allow you to make efficient use of high-strategic-taste advisors who don’t have good on-the-ground knowledge.

I’d be very excited to give Eliezer Yudkowsky at least a little bit of my regranting pot because anything that passes his incredibly tough sniff test maybe has a shot of working, but Eliezer does not spend his days hanging around the databases of career-transition-grantees or Manifund.
Conditional commitments that give Eliezer a fixed budget and the ability to write ideas that he wants to fund maximally take advantage of his eagle-eye view of the space, and general strategic taste.
More likely how this would work would be someone like me (a willing generalist, perhaps) interview him for an hour and extract all of them and turn them into 2-pagers, and then attach potential amounts to them and potential good founders / homes, go and pitch those people and institutions, and then get Eliezer to do a final sniff test.
This solves the information matching problem, where we channel Eliezer’s taste into concrete, dollar-attached projects, and he vibe-checks the initial conditions at the end. Funders could, if they wanted, demonstrate their relative confidence in different grantmakers by giving them different amounts to play with (this is notably much more flattering than “help us write our RFP.”)
Of course, apologies to non-Eliezer fans: sub-in your favorite, highly tasteful but busy strategic thinker.

This will get you better ideas than an open call or a proposal aggregation site. Many proposals cannot be written as multi-page grant applications, or humble Manifund posts.

Imagine asking the former national security advisor to apply for funding for his new national security think tank -- ridiculous! But, with conditional commitments, funders and their advisors can help make the initial conditions right.
Furthermore, many such grassroots proposals do not consider the full strategic picture and are best run by those already busy with full-time jobs, rather than a bright-eyed, agentic young person. And conditional commitments are a way to attract those eager generalists (generators?) with a “hey, this funder is interested in funding C proposal for $D amounts, but only if you make B conditions happen (eg. convince this person to join, or this institution to say yes)”, go wild.

A few wrinkles. How exactly do you make this commitment credible? In AI safety, we operate in a high-trust environment, but funders can be famously fickle. I’d recommend, at least to begin, shooting for explicit written confirmation somewhere -- an email, a google doc comment, a blogpost (but probably doesn’t have to be as uptight as a signed term sheet, lest it be subject to serious goodharting, reputation is a good motivator.) A big public commitment is another idea, where it can not only act as a commitment mechanism, but also a recruiting podcast. Folks should only quit their jobs after they pass the vibe check (and before most irreversible steps). And finally, don’t overcommit (but given the size of the funders’ budgets, this will probably not be a problem).

Just commit already

Giving away money is hard.

People have full-time jobs. Grantees really don’t make it easy to donate to them (where’s your donation email? what do you mean, my donation is too small for it to be worth your time?). And even still, to repatriate a patio-11-ism, the minimum rate of fraud isn’t zero.

One worry I have, as our third-wave of American philanthropists begin donating in earnest, is that this will not actually translate to more diversified taste. Why? I worry that these unimaginably busy AI employees and other-well-to-do with full-time jobs follow the advice of a trusted few who’ve done their research, defer to the regular months-long grant investigation processes by the usual suspects, or handwave their donation commitments to the future in favor of cute utilitarian arguments about “saving money to buy the galaxies” (more on this pernicious meme soon).

If done well, conditional commitments can diversify taste as well as an effective grantmaking program, attract world-class talent and enable unusual, effective projects. People only mobilize when there is money already on the table, and the openness of things that can be funded and the time-boundedness of commitments fights donor procrastination (sorry, future galaxies). If you’re interested in making one of these happen, I’ve got many ideas. Text me.

Thanks to Jesse Richardson for a lovely lunchtime conversation that inspired me to write this up.

The First Scattering

Conditional Commitments