This brings up something interesting: seems worthwhile to compare the internals of a ‘misgeneralizing,’ small n agent with those of large a n agents and check whether there seems to be a phase transition in how the network operates internally or not.
This brings up something interesting: seems worthwhile to compare the internals of a ‘misgeneralizing,’ small n agent with those of large a n agents and check whether there seems to be a phase transition in how the network operates internally or not.