Can we align automated alignment research? Currently building SOMA.
The interesting work lives at the intersection of mechanism design and multi-agent systems.