Forgetting Benchmarks for AI Companions: Testing Whether Old Context Actually Fades
A research framework for testing whether temporary, outdated and explicitly forgotten memories stop influencing future AI companion behavior.
12
A research framework for testing whether temporary, outdated and explicitly forgotten memories stop influencing future AI companion behavior.
A research framework for evaluating AI companion safety over weeks of interaction, where dependency, memory and personalization can accumulate.
A provenance model for tracing creator digital twin identity assets from source media and consent through model versions and public deployment.
A framework for evaluating whether gesture, posture, gaze and movement preserve a digital human’s identity across video interactions.
A research model for measuring mutual interest in AI-native social systems without collapsing relationships into a public popularity score.
A research framework for deciding which stored relationship memories should be retrieved into the current AI companion conversation.
How creator digital twins can communicate knowledge boundaries, uncertainty and escalation so fans know when they are talking to the AI rather than the human.
A benchmark framework for evaluating AI-native social discovery when a new user has no follow graph, history or established relationships.
How text sentiment, synthetic voice and avatar motion can be aligned so a digital human expresses one coherent emotional state.
A protocol for updating traits, preferences and behavior in persistent AI characters while preserving identity continuity.