arXiv:2510.04484v2 Announce Type: replace Abstract: The ability to control LLMs' emulated emotional states and personality traits is an essential step in enabling rich, human-centered interactions in socially interactive settings. We introduce PsySET, a Psychologically-informed benchmark to evaluate LLM Steering Effectiveness and Trustworthiness across the emotion and personality domains. Our study spans four models from different LLM families paired with various steering strategies, including prompting, fine-tuning, and representation engineering. Our results indicate that prompting is consis
The rapid advancement and deployment of LLMs necessitate a deeper understanding of their control and reliability, especially as they become more integrated into human-centric applications.
Controlling the psychological aspects of LLMs is critical for safe, effective, and trustworthy human-AI interaction across various applications, from customer service to social robotics.
The development of benchmarks like PsySET provides a standardized method to evaluate and compare psychological steering capabilities in LLMs, accelerating research and development in this area.
- · LLM developers
- · AI ethicists
- · Social robotics
- · Human-computer interaction research
- · Ineffectively steered LLM applications
- · Developers lacking steering expertise
Improved psychological steering will lead to more nuanced and context-aware AI interactions.
Enhanced control over LLM 'personalities' could open new avenues for personalized AI companions and therapeutic applications.
The ability to 'tune' AI psychology might lead to debates on the ethics of AI manipulation and the nature of digital influence.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.CL