The science of simulated users.
Questions, methods, and findings from the MatrAIx community—covering persona grounding, simulated users, interactive environments, evaluation, and identity drift.
Simulating the World with 8.3 Billion Personas
The MatrAIx technical report in plain language: a persona population, four kinds of studies, 1,010 reusable tasks, and 18,189 trials, with the demo video.
MatrAIx Core Team
Why Human Simulation Is the Next Step for AI Agents
Why AI needs human simulation: evidence, evaluation gaps, population models, individual digital twins, and a research agenda.
Xiaomin Li, Yuexing Hao, and Core Team
| Evaluation | Human | Simulation |
|---|---|---|
| 10K surveys | $28,560+ | $3.60–$104 |
| 1K chats | $8,568+ | $3–$87 |
| 100K outputs | $85,680 | $12.60–$365 |
The Cost of Synthetic-User Evaluation: Five Comparative Estimates
What realistic evaluation programs cost in human time and AI inference across surveys, chatbots, product testing, experiments, and annotation.
Xiaomin Li, Hanwen Xing, and Core Team
Measuring Persona Drift in Long-Horizon Agent Simulations
Measuring and correcting persona drift in long-horizon LLM agent interactions.
Sky Ng, Benjamin Cheng, and Core Team
Research
Have research or insights to share? We welcome community contributions.
Submit a post →