Gary Marcus Urges Scientific Approach to AI Alignment
Cognitive scientist questions understanding of new model's safety improvements.
TLDR
Gary Marcus posted that the world faces a nonobvious reason for deep trouble. He stated we are at the point where we need to "science the shit out of" alignment. Marcus noted that we do not know the first thing about how a new model like Astra works. He added that we also do not know how much of its putative improvements in alignment are due to benchmarkmaxxing or other factors.
Combined views
27K
1 Source, first seen 27d ago