Expert Personas Improve LLM Alignment but Damage Accuracy: Bootstrapping Intent-Based Persona Routing with PRISM
2026/03/19 by Zizhao Hu, Mohammad Rostami, Jesse Thomason · 9 voices · 1 citation
#cs.AI
paper · pdf
Abstract
Persona prompting can steer LLM generation towards a domain-specific tone and pattern. This behavior enables use cases in multi-agent systems where diverse interactions are crucial and human-centered tasks require high-level human alignment. Prior works provide mixed opinions on their utility: some report performance gains when using expert personas for certain domains and their contribution to data diversity in synthetic data creation, while others find near-zero or negative impact on general utility. To fully leverage the benefits of the LLM persona and avoid its harmfulness, a more comprehensive investigation of the mechanism is crucial. In this work, we study how model optimization, task type, prompt length, and placement can impact expert persona effectiveness across instruction-tuned and reasoning LLMs, and provide insight into conditions under which expert personas fail and succeed. Based on our findings, we developed a pipeline to fully leverage the benefits of an expert persona, named PRISM (Persona Routing via Intent-based Self-Modeling), which self-distills an intent-conditioned expert persona into a gated LoRA adapter through a bootstrapping process that requires no external data, models, or knowledge. PRISM enhances human preference and safety alignment on generative tasks while maintaining accuracy on discriminative tasks across all models, with minimal memory and computing overhead.
Citations
Cited by
Discussions
- It's been proven that telling an AI it's an expert in any domain makes it less accurate (see: arxiv.org/abs/2603.18507). Marc Andreesen is a fucking idiot, not an engineer. [bsky, 109 points, 9 comments]
- 2/2 Expert personas can meaningfully improve alignment with user intent and safety goals, but this often comes at the cost of reduced accuracy on core tasks. arxiv.org/abs/2603.18507 [bsky, 4 points, 0 comments]
- The fun part here is that expert prompts make models hallucinate more arxiv.org/abs/2603.18507 [bsky, 3 points, 0 comments]
- Macht sogar innerhalb von LLMs einen Unterschied, ob das Prompt ihnen die Rolle eines Experten zuschreibt: arxiv.org/abs/2603.18507 [bsky, 2 points, 0 comments]
- Your personal ignorance of the science actually undermines the point you are looking to make. Real research explains the phenomenon, but also shows it harms accuracy: arxiv.org/abs/2603.18507 [bsky, 1 points, 1 comments]
- Expert Personas Improve LLM Alignment but Damage Accuracy [hn, 1 points, 0 comments]
- Expert Personas Improve LLM Alignment but Damage Accuracy [hn, 1 points, 0 comments]
- Asking AI to act like an expert can make it less reliable arxiv.org/abs/2603.18507 [bsky, 0 points, 0 comments]
- arxiv.org/abs/2603.18507 Expert Personas Improve LLM Alignment but Damage Accuracy: Bootstrapping Intent-Based Persona Ro... [bsky, 0 points, 0 comments]
Related