<ul><li>Midjourney, known for its AI image generators, released new research on training text-based large language models (LLMs) to write more creatively.</li><li>The collaboration with New York University introduces two new techniques, DDPO and DORPO, to expand the range of possible outputs while maintaining coherence and readability.</li><li>The research goes beyond academic exercises and could fuel a new wave of LLM training among enterprise AI teams, product developers, and content creators.</li><li>By incorporating deviation, the models learn to produce high-quality but more varied responses, ensuring AI-generated stories explore a wider range of characters, settings, and themes.</li></ul>

Midjourney’s surprise: new research on making LLMs write more creatively

Discover more