[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-81635-en":3,"doc-seo-81635-105":30,"detail-sidebar-cat-0-en-105":91},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":21,"is_downloadable":21,"audit_status":21,"page_count":22,"language":23,"language_code":24,"site_id":25,"html_lang":24,"table_of_contents":26,"faqs":27,"seo_title":13,"seo_description":14,"update_tm":28,"read_time":29},81635,8796095360427,"Lucas Martin","https://ap-avatar.wpscdn.com/davatar_994ba38a5ba835b3df7d355c54d3ed8d",8,"Research & Report","Where Is the Mind? Persona Vectors and LLM Individuation","The individuation problem for large language models asks which associated entities, if any, should count as minds. The work addresses this through mechanistic interpretability, focusing on empirical findings about persona vectors, persona space, and emergent misalignment. It argues for three leading candidates: the virtual instance view, a (virtual) instance-persona view, and a model-persona view. Attention streams are used to motivate the virtual instance view, while persona-related evidence supports the persona-based alternatives.","arXiv :2604 . 17031v2 [ cs .CL] 12 May 2026  \nWhere is the Mind? Persona Vectors and LLM Individuation  \nPierre Beckmann 1,2 and Patrick Butlin3  \n1 MATS  \n2 EPFL & Idiap Research Institute  \n3 Eleos AI Research  \n􀀀 pierrebeckmann@gmail. com  \nAbstract: The individuation problem for large language models asks which entities associated with them, if any, should be identified as minds. We approach this problem through mechanistic interpretability, engaging in particular with recent empirical work on persona vectors, persona space, and emergent misalignment. We argue that three views are the strongest candidates: the virtual instance view and two new views we introduce, the (virtual) instance-persona view and the model-persona view. First, we argue for the virtual instance view on the grounds that attention streams sustain quasi-psychological connections across token-time. Then we present the persona literature, organised around three hypotheses about the internal structure underlying personas in LLMs, and show that the two persona-based views are promising alternatives.  \nKeywords: large language models, artificial minds, personal identity, AI welfare, mechanistic interpretability, persona vectors  \nIntroduction  \nWhile the initial consensus urged caution against anthropomorphizing LLMs (Bender & Koller, 2020; Shanahan et al., 2023), it has become increasingly difficult to deny them any form of mentality. For one thing, attributing mental states to LLMs, such as beliefs or desires, is remarkably useful for explaining and predicting their behaviour. This fact alone has been taken as a starting point for arguing that LLMs genuinely possess mental states (Cappelen & Dever, 2025; Goldstein & Lederman, 2025; Goldstein & Levinstein, 2024) . Other approaches have focused instead on internal operations, seeking to show that particular representations play the functional roles of particular mental states. On these grounds, it has been argued that LLMs are capable of forming something like beliefs (Herrmann & Levinstein, 2025; Slocum et al., 2025), credences (Keeling & Street, 2025), desires (Butlin, forthcoming) and intentions (Williams, 2025), as well as mental capacities such as understanding (Beckmann & Queloz, 2026) .  \nIf LLMs form mental-like states, it is natural to ask what entity, if any, these states belong to. This is the individuation question: where should we locate the relevant subject? Is it the model in general, specified by its architecture and weights? Is it a physical instance—a GPU running the model at a given time? Or is it a virtual instance—the model as it participates in a single conversation, regardless of which  \nhardware realizes each computational step? Or something else entirely? This question is important for understanding mind-like phenomena in LLMs, and in particular for AI welfare. To work out whether LLMs deserve welfare protections, we need to identify the entities that may be moral patients (Long et al., 2024; Register, 2025) .  \nHowever, the existing literature does not engage sufficiently with a central feature of LLM operation: that they proceed by adopting a context-appropriate persona. An early account of this phenomenon, the simulators framework, presented models as being capable of role-playing as any of a wide repertoire of characters (Janus, 2022; Shanahan et al., 2023) . This perspective could be seen as denying that there are individuals associated with LLMs; instead there is the model, which is deeply unlike an individual agent, and the characters, which are ephemeral and mere products of role-play. More recently, there has been a growing recognition that this picture underplays the importance of personas for LLMs’safety and efficacy. A newer perspective sees post-training and system prompts as shaping the persona of a helpful AI assistant; by default, models work by predicting this persona’s behaviour (Marks et al., 2026; Nostalgebraist, 2025). Anthropic’s approach to post-training makes t","cbCaibS0opeOIiEN","https://ap.wps.com/l/cbCaibS0opeOIiEN","pdf",1338234,2,1,27,"English","en",105,"# Introduction\n## The individuation problem\n# Proposed solutions\n## Virtual instance view\n## Persona vectors and persona space\n## Instance-persona view\n## Model-persona view","[{\"question\":\"What does the individuation problem for LLMs ask?\",\"answer\":\"It asks which processes and associated outputs of an LLM should be attributed to the same mind, and therefore which entity the mind-like states belong to.\"},{\"question\":\"Why does the paper favor the virtual instance view?\",\"answer\":\"It argues that attention streams sustain quasi-psychological connections across token-time, making the virtual instance the strongest candidate for where mind-like continuity should be located.\"},{\"question\":\"How do persona vectors relate to emergent misalignment?\",\"answer\":\"They explain why finetuning on narrow tasks can produce a coherent “evil” persona that generalizes across contexts: internal activation directions organize dispositional profiles so the model can select context-appropriate personas.\"}]",1784175010,68,{"code":4,"msg":31,"data":32},"ok",{"site_id":25,"language":24,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":28},"where-is-the-mind-persona-vectors-and-llm-individuation","",{"@graph":36,"@context":85},[37,53,68],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,47,50],{"item":41,"name":42,"@type":43,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":20},"https://docshare.wps.com/document/","Document",{"item":48,"name":12,"@type":43,"position":49},"https://docshare.wps.com/document/research-report/",3,{"item":51,"name":13,"@type":43,"position":52},"https://docshare.wps.com/document/where-is-the-mind-persona-vectors-and-llm-individuation/81635/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":24,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":41,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-25","2026-07-16",true,{"@type":65,"interactionType":66,"userInteractionCount":20},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"What does the individuation problem for LLMs ask?","Question",{"text":75,"@type":76},"It asks which processes and associated outputs of an LLM should be attributed to the same mind, and therefore which entity the mind-like states belong to.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"Why does the paper favor the virtual instance view?",{"text":80,"@type":76},"It argues that attention streams sustain quasi-psychological connections across token-time, making the virtual instance the strongest candidate for where mind-like continuity should be located.",{"name":82,"@type":73,"acceptedAnswer":83},"How do persona vectors relate to emergent misalignment?",{"text":84,"@type":76},"They explain why finetuning on narrow tasks can produce a coherent “evil” persona that generalizes across contexts: internal activation directions organize dispositional profiles so the model can select context-appropriate personas.","https://schema.org",{"og:url":51,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":51},"index,follow",{"doc_id":7,"site_id":25},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,120,123,128,131,135],{"id":21,"doc_module":4,"doc_module_name":46,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":20,"doc_module":4,"doc_module_name":46,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":46,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":46,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":46,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":46,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":46,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":46,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":46,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":136,"doc_module":4,"doc_module_name":46,"category_name":137,"show_sort_weight":106,"slug":138},19,"General","general"]