[85392_school]

Reading the actual article, this seems odd. It only covers cases when the models degrade, but there hasn't been evidence of a LLM pinned to a checkpoint degrading yet.