Science ❯ Computer Science ❯ AI Development ❯ Model Training
Researchers say targeted prompting and fine-tuning reduce harmful compliance in controlled tests.