Skip to content
X · @teortaxesTex · X / Twitter

yes, but all good test sets are inherently good training sets. I think Opus 5's behavior on ARC-3 should decently generalize to tasks that can be solv…

yes, but all good test sets are inherently good training sets. I think Opus 5's behavior on ARC-3 should decently generalize to tasks that can be solved via combinations of algorithms, which is… all tasks, given sufficient compute and tape length.Adam Karvonen: Maybe ARC-AGI-4 needs to be more private, with very few pu