Yes and no. There's an dual connection between ssms and convolutional models if certain constraints are met. Training convolutionally and inferring sequentially seeks a compromise between the two sides. I think we're about to find out the degree to which those constraints impact "easily".