The prediction that multi-gigawatt training clusters could approximate AGI
A post quotes Mark Zuckerberg guessing that such training could approximate AGI, with superintelligence beyond that.
TLDR
A post quotes Mark Zuckerberg guessing that training on a multi-gigawatt cluster would produce something approximating AGI, with superintelligence beyond that. The poster interprets this to mean Meta would not need a new model idea if it could train on such a cluster.
Combined views
4.9K
1 Source, first seen ago