Could AI Safety Make AI Less Safe?
Discussing the idea that interventions designed to prevent one type of AI harm can inadvertently make another worse.
Read post 11 min readWriting
Blog
A space for my thoughts on research, AI safety, and other things I find important.
The Future of Mechanistic Interpretability in Alignment
If a complete mechanistic explanation of a model is unrealistic, does mech interp still provide value for safety and alignment?
Read post 15 min read