I'm working on this project as well as Buck and Nate at Redwood Research. Fabien Roger might also collaborate.

The plan is for this to be a somewhat short project where we get pretty quick feedback (after less than 1 serial week and less than about 2 person weeks).
The value might be produced via a combination of us having better takes on alignment and this resulting in alignment people having better takes.

I might describe this project in more detail at a later point.

This is resolved based on ex-post assessment, so it will include unforeseeable factors. This includes accounting for opportunity cost.