> I am open to the argument

we agree then, that is the entirety of my argument. Getting a deep net especially one that is anywhere near even SLM size to be calibrated is tough, especially across domains. They claim calibration across a variety of datasets which is interesting.