Starling is a large language model trained by reinforcement learning from AI feedback focused on improving chatbot helpfulness.