https://www.landofgpt.com/mit-researchers-make-language-models-scalable-self-learners/