[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-seo-293634-105":3,"detail-sidebar-cat-0-en-105":81,"doc-detail-293634-en":131},{"code":4,"msg":5,"data":6},0,"ok",{"site_id":7,"language":8,"slug":9,"title":10,"keywords":11,"description":12,"schema_data":13,"social_meta":74,"head_meta":76,"extra_data":78,"updated_unix":80},105,"en","neuron-based-personality-trait-induction-in-large-language-models-poster-iclr-2025","Neuron-based Personality Trait Induction in Large Language Models - Poster - ICLR 2025","","Neuron-based Personality Trait Induction (NPTI) proposes a neuron-level approach to induce personality traits in large language models. It constructs PersonalityBench by generating Big Five descriptions and 180,000 situational questions derived from IPIP-NEO-300, then refines them to reduce bias. NPTI identifies trait-linked neurons via activation probability differences and modifies neuron behavior to increase positive aspects while deactivating negative ones. Experiments compare against prompt and training baselines and report automatic and human evaluation results on LLaMA-3-8B-Instruct.",{"@graph":14,"@context":73},[15,34,56],{"@type":16,"itemListElement":17},"BreadcrumbList",[18,23,27,31],{"item":19,"name":20,"@type":21,"position":22},"https://docshare.wps.com","Home","ListItem",1,{"item":24,"name":25,"@type":21,"position":26},"https://docshare.wps.com/document/","Document",2,{"item":28,"name":29,"@type":21,"position":30},"https://docshare.wps.com/document/research-report/","Research & Report",3,{"item":32,"name":10,"@type":21,"position":33},"https://docshare.wps.com/document/neuron-based-personality-trait-induction-in-large-language-models-poster-iclr-2025/293634/",4,{"url":32,"name":10,"@type":35,"image":36,"author":41,"headline":10,"publisher":44,"fileFormat":47,"inLanguage":8,"description":12,"dateModified":48,"datePublished":49,"encodingFormat":47,"isAccessibleForFree":50,"interactionStatistic":51},"DigitalDocument",{"url":37,"@type":38,"width":39,"height":40},"https://docshare.wps.com/thumbnails/neuron-based-personality-trait-induction-in-large-language-models-poster-iclr-2025/293634.png","ImageObject",300,407,{"name":42,"@type":43},"Bill Black","Person",{"url":19,"name":45,"@type":46},"DocShare","Organization","application/pdf","2026-11-01","2026-09-17",true,{"@type":52,"interactionType":53,"userInteractionCount":55},"InteractionCounter",{"@type":54},"ViewAction",25,{"@type":57,"mainEntity":58},"FAQPage",[59,65,69],{"name":60,"@type":61,"acceptedAnswer":62},"What problem does NPTI address in large language models?","Question",{"text":63,"@type":64},"It targets inducing personality traits in LLMs, addressing limitations of prompt-based induction instability and training-based resource costs.","Answer",{"name":66,"@type":61,"acceptedAnswer":67},"How is PersonalityBench constructed in NPTI?",{"text":68,"@type":64},"Personality descriptions for Big Five traits and facets are generated, and 180,000 situational questions are built from IPIP-NEO-300 with real-world topics, then refined to be objective and reduce bias.",{"name":70,"@type":61,"acceptedAnswer":71},"How does NPTI identify and modify personality-linked neurons?",{"text":72,"@type":64},"Neuron identification uses activation probability differences (δ) to classify neurons controlling positive vs. negative trait aspects. Neuron modification increases activation for positive neurons, deactivates negative neurons, and preserves neutral neuron values.","https://schema.org",{"og:url":32,"og:type":75,"og:title":10,"og:site_name":45,"og:description":12},"article",{"robots":77,"canonical":32},"index,follow",{"doc_id":79,"site_id":7},293634,1789666674,{"code":4,"msg":82,"data":83},"success",[84,88,92,96,101,106,111,115,120,123,127],{"id":22,"doc_module":4,"doc_module_name":25,"category_name":85,"show_sort_weight":86,"slug":87},"Story & Novel",90,"story-novel",{"id":26,"doc_module":4,"doc_module_name":25,"category_name":89,"show_sort_weight":90,"slug":91},"Literature",80,"literature",{"id":33,"doc_module":4,"doc_module_name":25,"category_name":93,"show_sort_weight":94,"slug":95},"Exam",70,"exam",{"id":97,"doc_module":4,"doc_module_name":25,"category_name":98,"show_sort_weight":99,"slug":100},5,"Comic",60,"comic",{"id":102,"doc_module":4,"doc_module_name":25,"category_name":103,"show_sort_weight":104,"slug":105},6,"Technology",50,"technology",{"id":107,"doc_module":4,"doc_module_name":25,"category_name":108,"show_sort_weight":109,"slug":110},7,"Healthcare",40,"healthcare",{"id":112,"doc_module":4,"doc_module_name":25,"category_name":29,"show_sort_weight":113,"slug":114},8,30,"research-report",{"id":116,"doc_module":4,"doc_module_name":25,"category_name":117,"show_sort_weight":118,"slug":119},9,"Religion & Spirituality",20,"religion-spirituality",{"id":118,"doc_module":4,"doc_module_name":25,"category_name":121,"show_sort_weight":118,"slug":122},"World Cup","world-cup",{"id":124,"doc_module":4,"doc_module_name":25,"category_name":125,"show_sort_weight":124,"slug":126},10,"Lifestyle","lifestyle",{"id":128,"doc_module":4,"doc_module_name":25,"category_name":129,"show_sort_weight":97,"slug":130},19,"General","general",{"code":4,"msg":82,"data":132},{"doc_id":79,"user_id":133,"nickname":42,"user_avatar":134,"doc_module":4,"category_id":112,"category_name":29,"doc_title":10,"doc_description":12,"doc_content":135,"file_id":136,"file_url":137,"file_type":138,"file_size":139,"view_count":55,"is_deleted":4,"is_public":22,"is_downloadable":22,"audit_status":22,"page_count":140,"language":141,"language_code":8,"site_id":7,"html_lang":8,"table_of_contents":142,"faqs":143,"seo_title":144,"seo_description":12,"update_tm":80,"read_time":145,"preview_page_count":4},24189269381491,"https://ap-avatar.wpscdn.com/avatar/160000cf11732dd8392?x-image-process=image/resize,m_fixed,w_180,h_180&k=1788146458752108895","Neuron-based Personality Trait Induction in Large Language Models  \nJia Deng, Tianyi Tang, Yanbin Yin, Wenhao Yang, Xin Zhao, Ji-Rong Wen  \n[2021200967@ruc.edu.cn](2021200967@ruc.edu.cn)  \nICLR 2025 Poster  \n [https://github.com/RUCAIBox/NPTI](https://github.com/RUCAIBox/NPTI)  \n01. Introduction  \n• Personality in LLMs  \n• Applications: personality simulation, sociological research. . .  \n• Assessment: MBTI and BFI, LLM as judge  \n• Current works  \n• Prompt based: explicit & implicit  \n• Training based: sft, rlhf  \n• Challenge  \n• Prompt-based induction: unstable, depending on the specific prompts .  \n• Model training: resource-intensive, requiring labeled data .  \nNeuron-based Personality Trait Induction in Large Language Models  \n01. Introduction  \nNPTI: Neuron-based approach for Personality Traits Induction in LLMs  \nNeuron-based Personality Trait Induction in Large Language Models  \n01. Introduction  \n• Contribution:  \n• PersonalityBench: personality descriptions, 180, 000 situational questions for BigFive traits .  \n• Neuron identification: method to locate trait-linked neurons .  \n• Neuron manipulation: method to modify personality neurons .  \n Neuron-based Personality Trait Induction in Large Language Models  \n02. Method-PersonalityBench Construction  \n• Personality Description Generation: Big Five traits and facets .  \n• Situational Question Construction: IPIP-NEO-300 questionnaire, real-world topics, refinement to ensure objectivity and minimize bias .  \nNeuron-based Personality Trait Induction in Large Language Models  \n02. Method-Neuron Identification  \n• Neuron Specification  \n• Probablity Calculation  \n• Neuron Identification  \nactivation value  \n• ℎ : output of the MHA module for a specific token  \n• σ: a non-linear activation function, such as SiLU  \n• 􀇁 1−3: learned projection matrices  \n• i: i-th neuron  \n• n: total number of generated tokens  \n• t: certain personality trait . A + represents positive aspect , and a represent the opposite  \n• δ: activation probability difference  \n• Neurons are classified as controlling the positive aspect of trait t if δ exceeds 10%, and vice versa .  \n Neuron-based Personality Trait Induction in Large Language Models  \n02. Method-Neuron Modification  \n• f: weighted function  \n• 􀇒 􀖕􀖡􀕽 : the original neuron value  \n• γ: a hyperparameter controlling the magnitude of change  \n• 􀇅 95 : the 9 5th percentile of the neuron's original activation  \n• Increase the activation value of positive neurons .  \n• Deactivate negative neurons .  \n• Maintain the original values of neutral neurons .  \n Neuron-based Personality Trait Induction in Large Language Models  \n03. Experiment-Settings  \n• Baseline methods:  \n• Simple prompt induction, P ^ 2 induction, PAS, ActAdd, SFT.  \n• Evaluation methods:  \n• Automatic Evaluation: SocialIQA, ChatGPT as evaluator, scored on a 1-5 scale .  \n• Human Evaluation: 200 questions, ranked by five judges .  \n Neuron-based Personality Trait Induction in Large Language Models  \n03. Experiment-Evaluation Results  \n• Automatic evaluation results on LLaMA-3-8B-Instruct model:  \n• Human evaluation results:  \nNeuron-based Personality Trait Induction in Large Language Models  \n03. Experiment-Evaluation Results  \n• Automatic evaluation results on LLaMA-3-8B-Instruct model:  \nhighest performance scores, lowest variance-> stable performance。  \n• Human evaluation results:  \nNeuron-based Personality Trait Induction in Large Language Models","cbCaioZqr0iBMtBu","https://ap.wps.com/l/cbCaioZqr0iBMtBu","pdf",1499389,23,"English","# 01. Introduction\n## Personality in LLMs\n## Applications and Assessment\n## Current Works and Challenges\n# 02. Method - PersonalityBench Construction\n## 02. Method - Neuron Identification\n## 02. Method - Neuron Modification\n# 03. Experiment - Settings\n## Baseline Methods\n## Evaluation Methods\n# 03. Experiment - Evaluation Results","[{\"question\":\"What problem does NPTI address in large language models?\",\"answer\":\"It targets inducing personality traits in LLMs, addressing limitations of prompt-based induction instability and training-based resource costs.\"},{\"question\":\"How is PersonalityBench constructed in NPTI?\",\"answer\":\"Personality descriptions for Big Five traits and facets are generated, and 180,000 situational questions are built from IPIP-NEO-300 with real-world topics, then refined to be objective and reduce bias.\"},{\"question\":\"How does NPTI identify and modify personality-linked neurons?\",\"answer\":\"Neuron identification uses activation probability differences (δ) to classify neurons controlling positive vs. negative trait aspects. Neuron modification increases activation for positive neurons, deactivates negative neurons, and preserves neutral neuron values.\"}]","Neuron-based Personality Trait Induction in Large Language Models - Poster - ICLR 2025 | PDF",58]