AI Safety Impact Markets (Release Notes v3.2 & v3.3)

In 2021, I, Dawn, wondered how I could contribute to making AI safer. I had donated to the Center on Long-Term Risk but I wanted to see whether I could do even better. I’m not a policy analyst or machine learning researcher, so what ended up at the top of my list was improving the funding of the AI safety ecosystem: impact markets. Now we’re a whole team working on realizing this vision!

There’s nothing about impact markets that makes them exclusive to AI safety. Speculative, early-stage projects exist in many spaces. But we worried that if we don’t focus on any particular group of users, none will think impact markets are for them.

So now we’re “AI Safety Impact Markets.” If you want to fundraise for something other than an AI safety project, you’re still welcome to do so on the platform, but we’ll mostly market it toward donors interested in AI safety, so other donors will be few and far in between.

Note that we went with “AI safety” because it seems to us to be the most inclusive term. We see it as a hypernym encompassing alignment research, research on making AI cooperative (and avoid s-risks), AI governance, and more.

Go to your profile, and you’ll find two new buttons:

1. Send me daily email notifications for new projects, and

2. Send me daily email notifications for new bounties.

Please switch them on to get an email digest (at most once a day) of new projects and bounties on the platform!

Our (current) top projects are now on the homepage!

Corollary: We’ve finally implemented sorting by the top donor “buy-in,” the support score. That means the projects that most donors with the highest scores have donated most to earliest bubble to the top. Want to express your support for a project? Donate early and register your donation!

We now have a ranking of the top contributors to a project. I went for the term “top contributor” to avoid confusion with “top donors,” because these top contributors are determined based on how much and how early they have donated to a project but not the retroactive evaluation of the project. That is, you can see this top list also on projects that are still ongoing and have not been evaluated.

You can now indicate your interest in a bounty with one click. That’s important so that others can decide whether they also want to work on the bounty. If the bounty says something like “10 submissions max,” and there are already 12 indications of interest, it’s starting to be questionable whether your submission will still be accepted. But if there are only 2 indications of interest, your chances are high!

(H/t Frankie)

It should be much easier now to find great tags for your project, at least within AI safety. We’ve added all the curate tags from the Alignment Forum!

I’ve also removed a bunch of code that related to impact certificates. The user profile, for example, still listed a user’s certificate transactions. Some people might’ve thought that “transactions” is a fancy term for donations and wondered whether someone with plenty of registered donations had no transactions listed. All that stuff is now gone, which should preempt such confusion.