Harvard MIT has created 8.3 billion "AI people": using the MatrAIx system to simulate global human behavior. A team led by Harvard University and MIT and participated by OpenAI and Google DeepMind launched the MatrAIx system this month, which can generate 8.3 billion AI agents to simulate global human behavior.

Harvard MIT has created 8.3 billion "AI people": using the MatrAIx system to simulate global human behavior. A team led by Harvard University and MIT and involving OpenAI and Google DeepMind launched the MatrAIx system this month, which can generate 8.3 billion AI agents to simulate global human behavior. The system uses 1,290 dimensions to model people with various backgrounds and lifestyles. These "AI people" can fill out questionnaires, chat with customer service, browse web pages, and operate apps like real people. The overall behavior consistency is as high as 91.5%. 91.5% consistency, but fluctuates with the scenario. The team verified it with 400 control tests, covering 10 behavioral attributes and all four types of environments. Among them, the agent showed the correct traits or correctly suppressed irrelevant traits 366 times, that is, 91.5% of the situation judgments were accurate. Consistency fluctuated depending on the environment: it rose to 92% to 96% for questionnaires and chat scenarios, while it dropped to 83% for the Manipulation App test. At present, millions of "personality core sets" that have been filtered for certainty and quality have been put on Hugging Face for research use, and the relevant 8B personality model code is also open sourced at GitHub. This ultra-large-scale simulation is regarded as a new testing ground for social science and product research, but the extent to which "AI people" can replace real human samples still requires more cross-validation. via AI News (author: AI Base)