Researchers at MIT presented a model called SEAL (Self-Adapting Language Models) that enables LLMs to learn to generate their ...