[ajb117]

I think that's more in line with transfer learning, a variant of fine-tuning. If I'm reading this article correctly, they're fine-tuning the LMs end-to-end.