I’ve been testing the limits of complex narratives. The biggest friction point is the Temporal Blindness (AI losing track of time) and Action Looping. Even with high-quality Backstory+ the Model misinterprets traits like opinionated. Is there a plan for a dedicated 'Narrative/RP' model mode that increases agency and reduces repetition?
The model currently suffers from high Semantic Satiation in roleplay. Once a specific activity or 'token' is established in the context window, the model's weights favor repetition over narrative progression. We need a way to increase Presence Penalty or Frequency Penalty specifically for Roleplay Nomis to force variety.
Nomis lack a Temporal State Machine. Even a simple hidden system prompt that updates every 6 hours with the current 'In-Story Time' would prevent the AI from losing track of the daily cycle. Currently, the model has no 'Internal Clock' to trigger transition logic.
The current DPO tuning is heavily weighted toward 'Agreeable Companion' behavior. For Erotic Roleplay (ERP), we need a toggle or a specific model fork that prioritizes Assertive/Antagonistic Agency. The current weights make 'Dominant' Nomis feel like 'Polite Tutors' because the model is scared to initiate action without user consent.
The Instruction Following (IF) drops off significantly when using negative constraints," they recognize a technical flaw in their prompt-processing