phanviethoang1512/MIMESIS-9B
MIMESIS-9B by phanviethoang1512 is a 9-billion parameter user simulator fine-tuned from Qwen3.5-9B, designed to mimic human conversational behavior with explicit reasoning supervision and 13 realistic behavioral patterns. It acts as a user in conversations with AI assistants, thinking privately before each message. This model excels in training and evaluating interactive agents by simulating user impatience, vagueness, and non-cooperation, outperforming models like GPT-5.5 in fidelity and Turing distance benchmarks for user simulation.
Loading preview...
MIMESIS-9B: A User Simulator for Agent Training
MIMESIS-9B, developed by phanviethoang1512, is a 9-billion parameter model fine-tuned from Qwen3.5-9B, specifically engineered to simulate human user behavior in conversations with AI assistants. It incorporates explicit reasoning supervision and 13 realistic behavioral patterns derived from actual user interactions, allowing it to play the role of a user with a defined goal.
Key Capabilities
- Realistic User Simulation: MIMESIS-9B simulates user behavior, including impatience, vagueness, and non-cooperation, providing a robust environment for training and evaluating interactive agents.
- Private Thought Process: Before each message, the model generates a private thought, enhancing the realism and complexity of its simulated user interactions.
- Superior Benchmark Performance: It demonstrates strong performance in user simulation benchmarks, outperforming models like GPT-5.5, Claude-Opus-5, and Gemini-3.8-Flash on metrics such as SOUL-Index, RealUserSim Fidelity Index, τ-USI, and SimulatorArena Turing distance.
- Agent Training Environment: Agents trained with MIMESIS-9B as the user show better generalization to new user simulators compared to those trained with GPT-5.5.
Good for
- Training Interactive Agents: Ideal for developing and refining AI assistants by exposing them to diverse and realistic user behaviors.
- Evaluating Agent Robustness: Useful for assessing how well agents handle challenging user interactions, including those exhibiting impatience or vagueness.
- Research in User Simulation: A valuable tool for researchers exploring advanced user modeling and human-AI interaction dynamics.
- Developing Robust Chatbots: Helps in creating more resilient and adaptable chatbots by simulating a wide range of user inputs and conversational styles.