E. Asgari et al. successfully reduced major errors in their AI system by refining prompts and workflows.