Part 6/12:
Moving beyond external scene descriptions, Shapiro explores modeling characters' internal states—interiority and stream-of-consciousness. Initial attempts involved rewriting scenes into first-person POV, a complex task for GPT-3, but through iterative prompt refinement, he manages to extract “interiority” data portraying characters' individual thoughts and motivations.
He emphasizes the importance of capturing character traits—beliefs, motives, preferences, backstories, and relationships—to enable the AI to simulate authentic internal dialogue. By translating external scenes into character-centric perspectives, the AI can generate “inner worlds” that inform believable reactions, decisions, and emotional states.