Evaluating Role-Consistency in LLMs for Counselor Training

TL;DR

This paper evaluates how well large language models maintain consistent roles in simulated online counseling scenarios, introducing an adversarial dataset and comparing multiple models' performance in role coherence.

Contribution

It introduces a new adversarial dataset and provides a comparative analysis of open-source LLMs' ability to sustain role consistency in virtual counseling simulations.

Findings

Vicuna model shows high role consistency

Adversarial attacks reveal vulnerabilities in LLMs

Open-source LLMs vary significantly in coherence

Abstract

The rise of online counseling services has highlighted the need for effective training methods for future counselors. This paper extends research on VirCo, a Virtual Client for Online Counseling, designed to complement traditional role-playing methods in academic training by simulating realistic client interactions. Building on previous work, we introduce a new dataset incorporating adversarial attacks to test the ability of large language models (LLMs) to maintain their assigned roles (role-consistency). The study focuses on evaluating the role consistency and coherence of the Vicuna model's responses, comparing these findings with earlier research. Additionally, we assess and compare various open-source LLMs for their performance in sustaining role consistency during virtual client interactions. Our contributions include creating an adversarial dataset, evaluating conversation coherence and persona consistency, and providing a comparative analysis of different LLMs.