https://www.corememory.com/p/exclusive-emmett-shear-is-back-with-softmax

This is obviously a very ambiguous and subjective question, I'd rather have an imperfect market than none

Close date extends indefinitely until the answer to this is clear.

"Central" is a high bar. Example: Inventing a better interpretability tool or a better method of reinforcement learning isn't enough to resolve this market YES, even if they're genuinely better than the alternatives and everyone adopts them. The most central YES resolution is is if their "organic alignment" works and is adopted by everyone, but it being a part of a future widely adopted alignment solution, or being widely accepted by AI safety as an important alignment innovation but not used, would work too. (This does rely on recognition, so if they do something objectively important but nobody other than them recognizes it, it still resolves no)