The unusually uncertain situation with AI makes this an unusually important time to donate to organizations which are in a position to influence how AI develops.
Here are my guesses as to which organizations can best use more funding. Everything in this area is changing rapidly, so this isn’t a prediction of where money will be needed 6 months from now.
Anima Labs is focused on understanding AIs through talking with them and caring about their welfare. Treating AIs with respect and understanding has important influence on how well AIs cooperate with us. It’s not hard to imagine a scenario where that makes a difference in whether the AIs that build superintelligence care about humans. Anima is underfunded due to a combination of most people neglecting this topic, and due to carelessly missing an SFF deadline.
The Verifiable Compute Foundation works on ways to verify a treaty to pause or pace AI development. See All hands on deck to build the datacenter lie detector for the general idea. It’s hard to tell whether they’re more constrained by funding or by talent.
Palisade Research is the organization I trust most to advise governments on what AI policies to adopt. The US government seems likely to soon adopt AI regulation in a hasty fashion that could easily be full of careless mistakes. Palisade combines a good understanding of AI with a good understanding of how government works. They’re fairly well funded, but more donations today could speed up their hiring plans.
The The AI Policy Network is somewhat like Palisade, but focused more on building influence with key members of Congress. As a 501(c)(4), they’re able to do lobbying that Palisade cannot.
If we can cause AIs to be corrigible, that reduces the risk of mistaken goals getting locked in forever. The Corrigibility Research Fund funds independent researchers who are working on making this reliable.
The AI Risk Mitigation Fund regrants donations to many independent researchers. I use it as a replacement for the Long-Term Future Fund.
Orthogonal is a strange group focused on a more rigorous long-term solution to AI than the others I’ve mentioned. It’s nearly broke, and doesn’t need a big budget – $200k per year would be plenty here, whereas the others that I’ve mentioned could maybe use a couple of million before seeing diminishing returns.
In many cases these endorsements are accidental byproducts of who I know well enough to trust. There are many more that I haven’t found ways to evaluate.
