​About the AI Research Circle

​The AI Research Circle is a community gathering at The Commons where we explore and discuss AI research papers together. You don’t need to be a researcher—just bring curiosity and an interest in the field.

​Each session, we choose a paper, break it down into plain language, and dive into open conversation. The goal is to make cutting-edge ideas accessible, spark thoughtful debate, and connect across disciplines.

​Session Details

​Paper: Persona Vectors: Monitoring and Controlling Character Traits in Language Models (Chen et al., 2025) 

​Large language models often present as a single “assistant” persona—but in practice, their personality can shift in surprising (and sometimes undesirable) ways due to prompting, fine-tuning, or training data. This paper introduces persona vectors: linear directions in a model’s activation space that correspond to specific character traits like evil, sycophancy, and hallucination propensity.

​We’ll use this paper as a jumping-off point to talk about what “personality” even means for LLMs, how linear directions emerge in activation space, and what this implies for alignment, safety, and tooling for model developers.

​Reading

​Primary (please read if you can):

• ​Persona Vectors: Monitoring and Controlling Character Traits in Language Models (Runjin Chen et al., 2025).

​Who should join

​Anyone interested in:

• ​How “personality” and “traits” show up in LLM behavior

• ​Mechanistic-ish tools for interpreting and steering models

• ​Practical alignment questions around fine-tuning, safety, and data pipelines

​No formal background in interpretability or alignment required—we’ll aim to keep things intuitive and conversational while still engaging for people who read a lot of papers.