MIT's new fine-tuning method lets LLMs learn new skills without losing old ones

Direct link to the paper for those who don't want to wade through the ads