The Dippy Roleplay subnet on Bittensor aims to create the world's best open-source roleplay LLM by leveraging the collective efforts of the open-source community. This subnet addresses the critical issue of loneliness, which affects a significant portion of the population and is linked to various mental and physical health problems.
Current SOTA LLMs (Claude, OpenAI etc.) are designed for the assistant use case and lack the empathetic qualities necessary for companionship. While some companies (c.ai, chai, inflection etc) have developed closed-source roleplay LLMs, the open-source alternatives lag significantly behind in performance.
Our team at Impel Intelligence Inc. knows this issue intimately through building Dippy, a proactive AI companion app for iOS. In this subnet, we will bring together the entire open-source eco-system to build the world's best roleplay LLM.
Roadmap
Given the complexity of creating a state of the art roleplay LLM, we plan to divide the process into 3 distinct phases.
Phase 1:
- Subnet launch with robust pipeline for roleplay LLM evaluation on public datasets and response length
- Public model leaderboard based on evaluation criteria
- Introduce Coherence and Creativity as a criteria for live model evaluation
Phase 2:
- Publicly release front-end powered by top miner submitted model of the week
- Segment model submission into different "expert" categories (funny, romantic, therapeutic etc)
- Models with the highest score in each personality type are chosen as "expert" models and made publicly available on the front-end
Phase 3:
- New Mixture of Experts model made as a baseline based on the "expert" models chosen from Phase 2
- Robust pipeline to evaluate new MOE model submissions against live evaluation criteria
- Expand the state of the art in roleplay LLMs through continuous iteration and data collection