Develops a Multi-Objective Direct Preference Optimization (MODPO) algorithm that can balance objectives of increased engagement for news content while maintaining editorial alignment without exacerbating polarization as baseline large language models do.
