Reaction
An AI interpretability finding and the case for incremental safety progress
A user argues that AI doomsayers can dismiss each advance as inadequate while overlooking their combined contribution to safety.
TLDR
A user points to a new AI interpretability finding while criticizing “doomers” who, they argue, can dismiss any single advance as insufficient. In their view, advances that fall short individually can still add up to keeping AI safe.
Combined views
—
2 Sources, first seen ago
