MatrAIx: Simulating World with 8.3 Billion Persona Agents

A new research paper on arXiv introduces MatrAIx, a population-scale simulated-user evaluation infrastructure designed to test AI systems and digital products with heterogeneous users. The system comprises three core components: Persona 8B, a dataset of 8.3 billion persona records represented by 1,290 categorical dimensions; the MatrAIx Playground, offering four environments (Survey, AI Chatbot, Web, App) for diverse user interactions; and a set of 1,010 evaluation tasks. The persona records are either sampled from a dependency graph preserving correlated attributes or derived from human-authored profiles. A quality-filtered coreset of approximately 1 million personas is released, including 599,847 human-grounded and 400,000 synthetic records. The paper, titled "MatrAIx: Simulating the World with 8.3 Billion Persona Agents," addresses the high cost and slow pace of human evaluation, proposing a scalable alternative that captures human diversity and interactive behavior. The infrastructure aims to improve offline evaluations by simulating heterogeneous user populations.