Improving Dynamic Object Interactions in Text-to-Video Generation with AI Feedback