Language models can also adopt personality archetypes. In a study led by Hang Jiang, a computer scientist at the Massachusetts Institute of Technology (MIT), researchers had GPT-3.5 assume hundreds of personas by prompting it to behave with different combinations of personality traits—for example, introverted, antagonistic, conscientious, neurotic, and closed to experience. For each persona, they had the model complete a standard personality test and write an 800-word childhood story that was then analyzed for psycholinguistic features associated with personality traits. The models dutifully manifested their assigned personalities both on the test and in the stories. Jiang says such models might allow researchers to test, say, how well people with different personalities would perform in various jobs.
https://media.mathstodon.xyz/media_attachments/files/110/710/550/612/553/068/original/1a90c6c97480d5f2.png