How To Clear The Talent Bottleneck

So it seems like smart and conscientious isn't enough.

Dan really thinks it is something to do with the metrics, and defining and finding competitive industries. A few points:

I am confused by how it seems like the average very competent PM at Meta TBD or xAI is perhaps heads and shoulders above the talent class that I normally work with. That's crazy, because I assume the talent class of my peers is pretty high. I wonder how they are able to get these PMs. I would like to meet this talent in industry, and see what the fire of shareholder value forges.

My naive take is that the difficulty in getting talent to do AI safety is that people aren't actually optimizing for goals, and so you'll want to essentially pick people out, and find them. Convincing them to do it with an offer that is competitive shouldn't be that hard. Just most people are low-agency, and the high-agency people don't wait for job offers that they apply to: the jobs apply to them.

CEOs are good at using their time wisely. So much of a CEOs time is spent doing talent-search. The top ones—Sam Altman, Mark Zuckerberg, Elon Musk—still find time to speak to potential recruits. And, as sama reminds us, they are efficient with their time.

It seems like you should hire someone who is well-connected and "cool" in a chief-of-staff type position to work on finding and competing for talent. They'll need to be pretty talented themselves, a founder-type perhaps. That's why the networks of the highest-status person in the organization should be wide, such that they can hire. The other people have more of a bottom-up type position, find the funnel, and reach out to anyone who shows just even a little bit of promise. These open-ended fellowships probably have incredible ROI.

It is easier to find talent if you can do a well-scoped-type thing. You have a project, with funding, with a good premise, with a vision. The problem is, for most projects that are that good, their founders tend to stay on. So you've, inherently, got the B-tier projects that need talent.

There is a chicken-egg problem where the best kinds of projects have money, talent, org buy-in, but you often need one of those factors in order to guarantee the other. You can solve this by having it be commonly known that the money is there. The fundamental thing for talented people is that they do need to believe in your vision, and they do need to believe that their unique talents can contribute in the upside.

That's why it is much more common for talented people to go bet on their own abilities in a start-up-type fashion, rather then earn some salary — that acts as more of an upper-bound to how much they can get out of their own position.

In some sense, the best people are the ones who want growth in their role to be something unlimitedly ambitious, and that it becomes much easier for them to move, because the opportunities stay. This is why the CAIS pitch was so compelling: it basically convinced me that there was an upper-bound and an inherent friction to getting what I want within the OP/Anthropic ecosystem, and that those limits would not apply to me at CAIS++, despite it being lower social status / lower-paid.

Get people with track records. Success begets success. Having people who are on an upwards trajectory with their careers, and who just win over and over again, are really tempting hires. This is the only way to separate the wheat from the chaff. The problem is, they need to feel like you are going to win, over and over again. So that's why money isn't enough. You need to have something that has momentum. You need to have a vision. That's why it makes sense for Mechanize and Leopold to be writing essays.

Okay so what are the practical takeaways:

The highest status person in your organization should be spending a good fraction of their time doing proactive hiring, and willing to drop everything to convince people. This is particularly true for the initialization. Get as good people as you can.

You not only need money, which is a static upper-bound on ambition, but a sense of momentum and an alignment on vision. People who win want to be on winning teams.

People are relatively sticky to their roles, but this is less true of ambitious people. With the right offer, they can be tempted away. They know their worth.

Particularly in non-competitive spaces, like nonprofits, you should try desperately from competitive spaces — companies, politics, attention-getting — and pay handsomely for that, for them to do something more amorphous. Founders make good hires.