The data-curation challenge for continual learning in AI

A user argues that current training methods cannot rapidly spread every possible skill across proprietary contexts if each must be deliberately post-trained. Continual learning might help, but if it also requires post-training, its data pipelines would need clean data about how models learn, not just clean question datasets. In an earlier critique, the user warned that more adaptable, brittle model weights could make existing LLM data-curation problems harder.