Superlinear - An EA Prize Platform

TLDR: At Nonlinear, we think bounties/prizes are underutilized. We’re trying an experiment to decentralize prize funding in the same way FTX decentralized grantmaking.

What would you do if you were a “prize regrantor” with $50,000 to fund your best x-risk prize ideas? If that excites you, apply here.

Are you a grantmaker with half-finished prize ideas? Let us know - we can help.

Why are we excited about bounties/prizes?

The nerd snipe formula: an interesting problem + a prize for solving it.

You’ll win $1m if you solve any of the seven Millennium Prize Problems. These problems nerd sniped a generation of math enthusiasts because the $1m prize says “this is an important problem!”

How successful have they been? One estimate is that ~100,000 people spent 10 hours trying to solve P=NP - that’s like 500 full-time PhD-level employees working for an entire year.

More examples: ELK, FLI Worldbuilding, Netflix Prize, crypto protocols pay out $2m-10m bug bounties (Gitcoin, Immunefi).

There are thousands of smart people in adjacent fields (e.g. capabilities researchers) who could be working with us overnight if we nerd sniped them with the right prize problems. Prizes are how we get them spending their nights and weekends working on reducing x-risk.

Prizes are scalable: orgs are bottlenecked by hiring and management, but prizes are only bottlenecked by imagination.

How big could this get?

Imagine you’re a grantmaker. You come across a spreadsheet of 100 prizes. You see a brilliant $1,000 prize idea, so you increase its value to $10,000. And repeat.

How many hours have been spent to date on AI alignment? Could we 100x that number with a big prize? Holden indicated he might be interested in funding a $1b prize for alignment. A $1b prize would be the biggest scientific prize of all time. It could boost the prestige of the field significantly because it’s a truly costly signal. If this is the most important problem in the world, shouldn’t it have the biggest prize?

What’s the P=NP, the Reimann hypothesis, or the Poincare conjecture of alignment? We don’t know, but we’d put out a bounty for anybody who can specify the $1b problem well enough to be bounty-able. Maybe we organize “prize hackathons” to generate ideas, etc.

---

If this vision excites you, join our discord and help us build it!

We think this could be big, but we're figuring it out as we go. Build with us!