Only in inference, but if you consider that they’re reinvesting inference performance in training I think the conflict argument is overblown.