Training Emergent Joint Associations: A Reinforcement Learning Approach to Creative Thinking in Language Models