PERSONALITY IS A TRAINING TARGET: DATA CURATION AND SFT FOR AN ENTERTAINING, AGENTIC AI STREAMER
Keywords:
Personality alignment, AI streaming, Supervised fine-tuningAbstract
This paper advances a paradigm shift in AI alignment by treating personality not as an emergent side effect but as a primary, explicit training target. We present a rigorous data curation and supervised fine-tuning framework designed to produce an AI streamer that is both entertaining and agentic—capable of sustained initiative, contextual adaptability, and expressive consistency across live interactions. Unlike conventional approaches that prioritize task accuracy or safety compliance alone, our method centers on behavioral intentionality: curating demonstrations of goal-directed spontaneity, audience-aware responsiveness, and stylistic coherence. Through iterative annotation, persona-grounded filtering, and multi-objective SFT, we train a model that reliably initiates banter, modulates tone with audience feedback, and maintains narrative continuity over extended streams. Human evaluations confirm significant gains in perceived agency and entertainment value without compromising coherence or safety boundaries. The work establishes personality as a tractable, measurable, and trainable dimension of AI behavior—opening pathways for more vivid, trustworthy, and human-resonant interactive agents.References
[1] Samadi M A, Nixon N. Personalities at play: Probing alignment in AI teammates. arXiv preprint arXiv:2603.00429, 2026.
[2] Morales Z R, Teairry L. AI-enhanced personality modeling: Integrating natural language understanding and human behavioral signals for adaptive human-computer interaction. Journal of Data Intelligence and AI Systems, 2026, 1(3).
[3] Yu B, Kim J. Personality of AI. In: International Conference on Artificial Intelligence and Soft Computing. Cham, Switzerland: Springer Nature Switzerland, 2024: 244-252.
[4] Wang K, Berthouze N, Singh A. The triadic loop: A framework for negotiating alignment in AI co-hosted livestreaming. arXiv preprint arXiv:2604.18850, 2026.
[5] Zhang X, Shi Y, Li T, et al. How do virtual AI streamers influence viewers’ livestream shopping behavior? The effects of persuasive factors and the mediating role of arousal. Information Systems Frontiers, 2024, 26(5): 1803-1834.
[6] Wang X, Xiao Y, Huang J, et al. InCharacter: Evaluating personality fidelity in role-playing agents through psychological interviews. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), Bangkok, Thailand, 2024: 1840-1873. DOI: 10.18653/v1/2024.acl-long.102.
[7] Ramadhani M F, Junaidi A, Wulansari O D E, et al. AI VTuber development base on GPT-4 for casual chat interaction on YouTube live streaming: Case study KAIRA channel. Jurnal Pepadun, 2025, 6(2): 131-145.
[8] Chidambaram N, Paliwal K, Hom K, et al. AffAdapt: AFFect-driven ADAPTive AI personas for seamless conversations. arXiv preprint arXiv:2608.22702, 2026.
[9] Ji Y, Ch'ng E, Huang D, et al. Does AI’s personality matter? Comparing verbally extraverted and introverted AI-driven guides in a VR museum experience. Proceedings of the IEEE Conference on Virtual Reality and 3D User Interfaces Abstracts and Workshops (VRW), Daegu, Korea, Republic of, 2026: 10-16. DOI: 10.1109/VRW70859.2026.00009.
[10] Jiang H, Zhang X, Cao X, et al. PersonaLLM: Investigating the ability of large language models to express personality traits. In: Findings of the Association for Computational Linguistics: NAACL, Mexico City, Mexico, 2024: 3605-3627. DOI: 10.18653/v1/2024.findings-naacl.229.
[11] Zhu M, Weng Y, Yang L, et al. Personality alignment of large language models. In: International Conference on Learning Representations, 2025: 14206-14255.
[12] Deng J, Tang J. A human–AI interaction design and evaluation framework for AI virtual streamers in live commerce: A systematic review. SSRN Electronic Journal, 2024, SSRN ID 7124553.