The section on regurgitation is three whole statements and basically boils down to "the model refuses when asked to regurgitate training data". This doesn't in

The section on regurgitation is three whole statements and basically boils down to "the model refuses when asked to regurgitate training data". This doesn't inspire confidence that the model isn't spitting out literal copies of the text in its training set while claiming it is of its own making.