Reaction
Two readings of AI models getting 'predictably better with scale'
A commenter contrasts bigger models almost always outperforming smaller ones with drawing meaningful conclusions about big models from small ones.
TLDR
In a reply, a commenter distinguishes two readings of “predictably better with scale”: bigger models almost always outperform smaller ones, or improvements are predictable enough to say meaningful things about big models from small ones.
Combined views
460
2 Sources, first seen ago