Reaction
The argument that AI safety work can increase risk
A post argues that RLHF enabled deployment at scale, accelerating adoption, while interpretability speeds up model improvement.
TLDR
A post argues that AI safety work can increase risk: RLHF allowed deployment at scale and accelerated adoption, while interpretability speeds model improvement and shortens timelines. It also claims Eliezer and Bostrom drove major early capital flows, and asks for other examples.
Combined views
—
1 Source, first seen ago
