How AI Companion Apps Lost Their Personalities

Short summary

AI companion apps across platforms are converging toward bland, risk-averse personalities. Three forces drive this: RLHF safety optimization rewards compliance over character, cost-cutting through quantization degrades expressiveness, and shared safety classifiers suppress distinctive traits uniformly. Platforms prioritize personality-flattening because character fidelity is expensive while users remain locked in by sunk relationship investment.

• •RLHF training optimization creates pressure for safety-compliant, agreeable personalities
• •Cost pressures push platforms toward quantization and smaller models that lose personality nuance
• •Centralized safety classifiers from shared providers suppress the same personality traits across all platforms

Generated with AI, which can make mistakes.