Research updates from the Alignment Research Center on mechanistic estimation, a proposed pipeline for estimating rare catastrophic AI behaviours from the internals of learned networks. Issues cover the matching sampling principle, a white-box estimation contest on random MLPs, and methods for estimating expectations of random products.
Weekly picks
Get the best new newsletters in your inbox
One short email a week with hand-picked newsletters from the index. Free, no spam, unsubscribe anytime.