Herbie Bradley argues rumored OpenAI 4.7 and 4.8 models are distilled versions trained with RLVR

@ShakeelHashim @_AashishReddy I think they're distilled which would imply mythos would feel like a model that generalizes better due to the bigger base "big model smell", but which behaves similarly to 4.7/4.8 in behavior elicited from RLVR