The first study investigates a method for modifying the personality of large language models (LLMs) using activation engineering. Researchers identified and manipulated activation vectors linked to specific personality traits, visualizing their effects and exploring ethical implications. The second study assesses LLMs' ability to diagnose conflict sources as stemming from differences in beliefs or values. Using vignettes, the researchers compared LLMs' performance to humans', finding that while LLMs understood the distinction, they exhibited biases in their diagnoses, particularly GPT-4 with a proximate scale. Both studies contribute to understanding LLM capabilities and the ethical considerations of their application.

2025

Large Language Model Diagnostic Competence in Conflict Mediation

TheMediator.AI is a new AI-powered platform designed to provide affordable and accessible online dispute resolution services.  Users can resolve personal conflicts conveniently through a mobile app for a low cost of $4.99 per dispute.  The platform ensures confidentiality and impartiality, using AI to analyze disputes and suggest resolutions.  The process involves both parties presenting their perspectives, answering questions (kept private from the other party), and receiving a proposed solution.  Data is deleted after 30 days, unless a PDF record of the resolution is generated.

Large Language Model Diagnostic Competence in Conflict Mediation

TheMediator

Related tracks