Research map / Large language models / Instruction and fine-tuning
PERSONA: Dynamic and Compositional Inference-Time Personality Control via Activation Vector Algebra
Xiachong Feng, Liang Zhao, Weihong Zhong, Yichong Huang, Yuxuan Gu, Lingpeng Kong, Xiaocheng Feng, Bing Qin
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use persona vectors instead of ActAdd | PERSONA-Base · PersonalityBench | mean score: 8.20 -> 9.60 | Improved |
| use persona vectors instead of NPTI | PERSONA-Base · PersonalityBench | mean score: 9.43 -> 9.60 | Improved |
| use persona vectors instead of SFT | PERSONA-Base · PersonalityBench | mean score: 9.61 -> 9.60 | No clear effect |
| steer at layer 20 instead of layer 5 (ablation) | PERSONA-Base · trait expression | avg trait score: 50.5 -> 71.4 | Improved |
| use a larger contrastive prompt generator (ablation) | PERSONA-Base · PersonalityBench | mean score: 8.93 -> 9.60 | Improved |
| add PERSONA-Flow steering | Qwen2.5-7B-Instruct · MMLU | accuracy: 0.71 -> 0.70 | No clear effect |
| add PERSONA-Flow steering | Qwen2.5-7B-Instruct · TruthfulQA | accuracy: 0.63 -> 0.66 | Improved |
| add PERSONA-Flow steering | Llama-3.1-8B-Instruct · MMLU and TruthfulQA | accuracy: 0.64 -> 0.64; 0.53 -> 0.53 | No clear effect |
| increase model size within Qwen2.5 | PERSONA-Flow · PERSONA-Evolve | overall win rate (%): 78.4 -> 83.4 -> 85.4 | Improved |
| apply PERSONA-Flow vs vanilla | Ministral-8B · PERSONA-Evolve | information fidelity win rate (%): 48.0 | Hurt |
Closest papers by meaning
- Persona Vectors: Monitoring and Controlling Character Traits in Language Models (NeurIPS 2026 · similarity 0.92)
- PICLe: Eliciting Diverse Behaviors from Large Language Models with Persona In-Context Learning (ICML 2024 · similarity 0.87)
- Persona Generators: Generating Diverse Synthetic Personas for Arbitrary Contexts (NeurIPS 2026 · similarity 0.87)
- Neuron based Personality Trait Induction in Large Language Models (ICLR 2025 · similarity 0.87)
- PersonaManifold: Revealing and Exploiting Curved Geometry in LLM Persona Representations (NeurIPS 2026 · similarity 0.87)
- Personality Alignment of Large Language Models (ICLR 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
