[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-0-en-105":3,"doc-seo-273542-105":59,"doc-detail-273542-en":131},{"code":4,"msg":5,"data":6},0,"success",[7,13,18,23,28,33,38,43,48,51,55],{"id":8,"doc_module":4,"doc_module_name":9,"category_name":10,"show_sort_weight":11,"slug":12},1,"Document","Story & Novel",90,"story-novel",{"id":14,"doc_module":4,"doc_module_name":9,"category_name":15,"show_sort_weight":16,"slug":17},2,"Literature",80,"literature",{"id":19,"doc_module":4,"doc_module_name":9,"category_name":20,"show_sort_weight":21,"slug":22},4,"Exam",70,"exam",{"id":24,"doc_module":4,"doc_module_name":9,"category_name":25,"show_sort_weight":26,"slug":27},5,"Comic",60,"comic",{"id":29,"doc_module":4,"doc_module_name":9,"category_name":30,"show_sort_weight":31,"slug":32},6,"Technology",50,"technology",{"id":34,"doc_module":4,"doc_module_name":9,"category_name":35,"show_sort_weight":36,"slug":37},7,"Healthcare",40,"healthcare",{"id":39,"doc_module":4,"doc_module_name":9,"category_name":40,"show_sort_weight":41,"slug":42},8,"Research & Report",30,"research-report",{"id":44,"doc_module":4,"doc_module_name":9,"category_name":45,"show_sort_weight":46,"slug":47},9,"Religion & Spirituality",20,"religion-spirituality",{"id":46,"doc_module":4,"doc_module_name":9,"category_name":49,"show_sort_weight":46,"slug":50},"World Cup","world-cup",{"id":52,"doc_module":4,"doc_module_name":9,"category_name":53,"show_sort_weight":52,"slug":54},10,"Lifestyle","lifestyle",{"id":56,"doc_module":4,"doc_module_name":9,"category_name":57,"show_sort_weight":24,"slug":58},19,"General","general",{"code":4,"msg":60,"data":61},"ok",{"site_id":62,"language":63,"slug":64,"title":65,"keywords":66,"description":67,"schema_data":68,"social_meta":124,"head_meta":126,"extra_data":128,"updated_unix":130},105,"en","interpretable-humans-alien-llms-expert-analysis-of-latent-structures-in-assessment-responses","Interpretable Humans - Alien LLMs - Expert Analysis of Latent Structures in Assessment Responses","","The evaluation of large language models depends on human-designed assessments and the implicit premise that AI and humans share underlying cognitive constructs. This work tests that premise by comparing latent factors extracted from human responses versus six LLMs on quantitative reasoning and chemistry assessments. Exploratory factor analysis yields separate factor structures for each group, then subject-matter experts interpret the resulting factor graphs blindly. Experts explain most human-derived factors, but interpret none of the quantitative-reasoning LLM factors and only half in chemistry, indicating opaque, non-human reasoning mechanisms.",{"@graph":69,"@context":123},[70,84,106],{"@type":71,"itemListElement":72},"BreadcrumbList",[73,77,79,82],{"item":74,"name":75,"@type":76,"position":8},"https://docshare.wps.com","Home","ListItem",{"item":78,"name":9,"@type":76,"position":14},"https://docshare.wps.com/document/",{"item":80,"name":40,"@type":76,"position":81},"https://docshare.wps.com/document/research-report/",3,{"item":83,"name":65,"@type":76,"position":19},"https://docshare.wps.com/document/interpretable-humans-alien-llms-expert-analysis-of-latent-structures-in-assessment-responses/273542/",{"url":83,"name":65,"@type":85,"image":86,"author":91,"headline":65,"publisher":94,"fileFormat":97,"inLanguage":63,"description":67,"dateModified":98,"datePublished":99,"encodingFormat":97,"isAccessibleForFree":100,"interactionStatistic":101},"DigitalDocument",{"url":87,"@type":88,"width":89,"height":90},"https://docshare.wps.com/thumbnails/interpretable-humans-alien-llms-expert-analysis-of-latent-structures-in-assessment-responses/273542.png","ImageObject",300,407,{"name":92,"@type":93},"Violet","Person",{"url":74,"name":95,"@type":96},"DocShare","Organization","application/pdf","2026-10-28","2026-09-15",true,{"@type":102,"interactionType":103,"userInteractionCount":105},"InteractionCounter",{"@type":104},"ViewAction",17,{"@type":107,"mainEntity":108},"FAQPage",[109,115,119],{"name":110,"@type":111,"acceptedAnswer":112},"What core assumption about LLM evaluation does the study challenge?","Question",{"text":113,"@type":114},"It challenges the assumption that LLM performance on human-designed assessments reflects the same underlying skills and constructs used by human learners, with similar item-to-skill mappings.","Answer",{"name":116,"@type":111,"acceptedAnswer":117},"How were latent factors extracted for humans and LLMs?",{"text":118,"@type":114},"The study applied Exploratory Factor Analysis (EFA) separately to human responses and to responses from six LLM versions for the same chemistry and quantitative reasoning assessments.",{"name":120,"@type":111,"acceptedAnswer":121},"How did subject-matter experts interpret the factors?",{"text":122,"@type":114},"Subject-matter experts reviewed the factor graphs blindly, without knowing whether a factor came from humans or LLMs, and then assigned pedagogical meaning to the constructs.","https://schema.org",{"og:url":83,"og:type":125,"og:title":65,"og:site_name":95,"og:description":67},"article",{"robots":127,"canonical":83},"index,follow",{"doc_id":129,"site_id":62},273542,1789471981,{"code":4,"msg":5,"data":132},{"doc_id":129,"user_id":133,"nickname":92,"user_avatar":134,"doc_module":4,"category_id":39,"category_name":40,"doc_title":65,"doc_description":67,"doc_content":135,"file_id":136,"file_url":137,"file_type":138,"file_size":139,"view_count":105,"is_deleted":4,"is_public":8,"is_downloadable":8,"audit_status":8,"page_count":140,"language":141,"language_code":63,"site_id":62,"html_lang":63,"table_of_contents":142,"faqs":143,"seo_title":144,"seo_description":67,"update_tm":130,"read_time":145,"preview_page_count":4},4398048950312,"https://ap-avatar.wpscdn.com/avatar/400002538284de19e3c?_k=1778320343897328908","Interpretable Humans, Alien LLMs: Expert Analysis of Latent Structures  \nin Assessment Responses*  \nAlona Strugatski∗ , Licol Zeinfeld∗ , Jason Cooper,  \nShelley Rap, Gil Schwarts, Giora Alexandron  \nWeizmann Institute of Science  \n∗Equal contribution  \n{alona.faktor, licol.zeinfeld, jason.cooper, shelley.rap, gil.schwarts, [giora.alexandron@weizmann.ac.il}](giora.alexandron@weizmann.ac.il})  \narXiv : 2608 . 178 10v 1 [cs .CL] 18 Aug 2026  \nAbstract  \nThe evaluation of large language models (LLMs) relies heavily on human-designed assessments, implicitly assuming that AI and humans employ similar underlying cognitive constructs. Challenging this assumption, we investigate whether the latent factors governing LLM performance carry the same substantive, human-interpretable meaning as the cognitive constructs governing human learners. Using responses from humans and six LLMs across quantitative reasoning and chemistry assessments, we conducted Exploratory Factor Analysis (EFA) separately for both groups. SubjectMatter Experts (SMEs) then blindly evaluated the resulting factor graphs to ascribe pedagogical meaning to the emerged constructs. SMEs successfully interpreted most of the humanderived factors. Conversely, they could not ascribe meaning to any LLM-derived factors in quantitative reasoning and interpreted only half of the LLM factors in chemistry. By combining data-driven EFA with blind expert interpretation, this framework shows that LLMs frequently operate on statistically opaque mechanisms distinct from human reasoning.  \n1 Introduction  \nAs large language models (LLMs) are increasingly applied to a wide range of cognitive tasks, the systematic evaluation of their capabilities and limitations has become increasingly important (Raiaan et al., 2024 ; Laskar et al., 2024) . A common approach to LLM evaluation is using established assessment instruments, particularly standardized exams that were developed for measuring human learners (Achiam et al., 2023 ; Zhong et al., 2024) . Moreover, such evaluations typically follow the rationale that the assessment that was used to evaluate human skills in a certain domain can be used to make arguments about LLM skills within that domain (Katz et al., 2024 ; Sánchez Salido et al., 2025 ;  \n*Accepted for publication at AIME 2026 .  \nJimenez et al., 2023 ; Borges et al., 2024 ; Yacobsonet al., 2026 ; Chlapanis et al., 2025) . However, this seems LLMs’ to make an implicit assumption that LLMs performance on those tests can be attributed to the same skills that humans apply to solve the test items, and that the items-to-skill mapping is similar for humans and LLMs.  \nRecent works start to challenge these LLM evaluation practices from educational measurement perspectives, showing, for example, that the structure offactors differs significantly for LLMs and human learners (Strugatski et al., 2026) . This connects to content-driven analyses that reported that LLMand humans are impacted differently from task dimensions (Zeinfeld et al., 2026), and IRT-driven analyses showing that LLM response patterns on educational assessments deviate from the norm and appear as aberrant, compared to human learners (Strugatski and Alexandron, 2025 ; Sorenson and Hanson, 2024) . This raises questions on what are the factors that govern LLM performance.  \nFor human respondents, factors are often interpreted in relation to established theories, domain knowledge, and expected response processes (Flora and Flake, 2017 ; Leighton and Gierl, 2007) . For LLMs, however, this issue remains under-examined in the literature and is of interest both for understanding LLM capabilities and for designing assessments for contexts in which respondents may use AI (legitimately or not) . Recent work has analyzed LLM response datasets to identify factors that emerge from their response patterns (Maimonet al., 2025) . However, the identified factors were not directly compared with the factors that explain human performance on the s","cbCaikgjJRCGK8KH","https://ap.wps.com/l/cbCaikgjJRCGK8KH","pdf",4714836,11,"English","# Abstract\n# 1 Introduction\n## Methodological Flowchart\n## Research question and approach","[{\"question\":\"What core assumption about LLM evaluation does the study challenge?\",\"answer\":\"It challenges the assumption that LLM performance on human-designed assessments reflects the same underlying skills and constructs used by human learners, with similar item-to-skill mappings.\"},{\"question\":\"How were latent factors extracted for humans and LLMs?\",\"answer\":\"The study applied Exploratory Factor Analysis (EFA) separately to human responses and to responses from six LLM versions for the same chemistry and quantitative reasoning assessments.\"},{\"question\":\"How did subject-matter experts interpret the factors?\",\"answer\":\"Subject-matter experts reviewed the factor graphs blindly, without knowing whether a factor came from humans or LLMs, and then assigned pedagogical meaning to the constructs.\"}]","Interpretable Humans - Alien LLMs - Expert Analysis of Latent Structures in Assessment Responses | PDF",28]