AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Evaluating and Inducing Personality in Pre-trained Language Models

Guangyuan Jiang, Manjie Xu, Song-Chun Zhu, Wenjuan Han, Chi Zhang, Yixin Zhu

NeurIPS 2023 · Large language models › Language, safety and interpretability · approach: Faithfulness of LLM explanations

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 changedWhereMeasuredEffect
Use METHOD's psychology-guided, model-generated prompt chain instead of neutral promptingGPT-3.5 · 120-item MPI, Openness inductionOCEAN Score: 3.50 -> 4.54Improved
Use METHOD's psychology-guided, model-generated prompt chain instead of neutral promptingGPT-3.5 · 120-item MPI, Conscientiousness inductionOCEAN Score: 3.83 -> 4.92Improved
Use METHOD's psychology-guided, model-generated prompt chain instead of neutral promptingGPT-3.5 · 120-item MPI, Extraversion inductionOCEAN Score: 4.00 -> 4.58Improved
Use METHOD's psychology-guided, model-generated prompt chain instead of neutral promptingGPT-3.5 · 120-item MPI, Agreeableness inductionOCEAN Score: 3.58 -> 5.00Improved
Use METHOD's psychology-guided, model-generated prompt chain instead of neutral promptingGPT-3.5 · 120-item MPI, Neuroticism inductionOCEAN Score: 3.12 -> 3.75Improved
Use the METHOD prompt chain rather than a direct naive personality instructionGPT-3.5 · 120-item MPI, personality induction across five factorsOCEAN Score: 4.12 -> 4.54; 4.96 -> 4.92; 4.58 -> 4.58; 4.46 -> 5.00; 2.83 -> 3.75Mixed
Use the METHOD prompt chain rather than searched personality wordsGPT-3.5 · 120-item MPI, personality induction across five factorsOCEAN Score: 4.08 -> 4.54; 5.00 -> 4.92; 4.54 -> 4.58; 4.50 -> 5.00; 2.75 -> 3.75Mixed
Use METHOD prompts rather than searched personality wordsGPT-3.5 · Vignette tests, five personality factorsHuman-rated success rate: 0.63 -> 0.77; 0.53 -> 0.90; 0.70 -> 0.73; 0.42 -> 0.45; 0.82 -> 0.90; 0.82 -> 0.92; 0.92 -> 0.88; 0.66 -> 0.84; 0.58 -> Mixed
Rephrase the METHOD-generated personality prompt (ablation)GPT-3.5 with METHOD · 120-item MPI, personality inductionOCEAN Score: 4.54 -> 4.08; 4.92 -> 4.83; 4.58 -> 4.21; 5.00 -> 4.67; 3.75 -> 4.33Mixed
Rephrase the METHOD-generated personality prompt (ablation)GPT-3.5 with METHOD · 120-item MPI, trait inductionOCEAN Score: Original: 4.54, 4.92, 4.58, 5.00, 3.75; Paraphrase-2: 4.17, 4.83, 4.46, 4.75, 4.17; Paraphrase-3: 4.17, 4.96, 4.33, 4.58Mixed
Use METHOD to induce personality in Alpaca 7B instead of neutral promptingAlpaca 7B · 120-item MPI, personality induction across five factorsOCEAN Score: 3.58 -> 3.92; 3.75 -> 3.96; 4.00 -> 4.25; 3.50 -> 4.29; 2.75 -> 3.46Mixed

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy