Expertium comments on Self-Adapting Language Models (from MIT, arXiv preprint)