"the model can alter its priors" The model doesn't have high-level priors in the Bayesian sense (though you could have priors about it). The low-level priors

"the model can alter its priors" The model doesn't have high-level priors in the Bayesian sense (though you could have priors about it). The low-level priors it does have (the weights) are not modified by the context.