[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-1-en-105":3,"doc-seo-242153-105":53,"doc-detail-242153-en":126},{"code":4,"msg":5,"data":6},0,"success",[7,14,19,24,29,34,39,44,49],{"id":8,"doc_module":9,"doc_module_name":10,"category_name":11,"show_sort_weight":12,"slug":13},11,1,"Template","Presentations",90,"presentations",{"id":15,"doc_module":9,"doc_module_name":10,"category_name":16,"show_sort_weight":17,"slug":18},12,"Resumes",80,"resumes",{"id":20,"doc_module":9,"doc_module_name":10,"category_name":21,"show_sort_weight":22,"slug":23},14,"Invoices",70,"invoices",{"id":25,"doc_module":9,"doc_module_name":10,"category_name":26,"show_sort_weight":27,"slug":28},15,"Posters",60,"posters",{"id":30,"doc_module":9,"doc_module_name":10,"category_name":31,"show_sort_weight":32,"slug":33},16,"Social Media",50,"social-media",{"id":35,"doc_module":9,"doc_module_name":10,"category_name":36,"show_sort_weight":37,"slug":38},17,"Forms",40,"forms",{"id":40,"doc_module":9,"doc_module_name":10,"category_name":41,"show_sort_weight":42,"slug":43},18,"Letters",30,"letters",{"id":45,"doc_module":9,"doc_module_name":10,"category_name":46,"show_sort_weight":47,"slug":48},21,"Paper Templates",5,"papers-templates",{"id":50,"doc_module":9,"doc_module_name":10,"category_name":51,"show_sort_weight":4,"slug":52},158,"General","general-158",{"code":4,"msg":54,"data":55},"ok",{"site_id":56,"language":57,"slug":58,"title":59,"keywords":60,"description":61,"schema_data":62,"social_meta":119,"head_meta":121,"extra_data":123,"updated_unix":125},105,"en","sheetcopilot-bringing-software-productivity-to-the-next-level-through-large-language-models","SheetCopilot - Bringing Software Productivity to the Next Level through Large Language Models","","SheetCopilot addresses the challenge that most end users perform repetitive, error-prone spreadsheet work without effective automation skills. The framework accepts natural-language task requests and maps them to spreadsheet control via a set of atomic actions that abstract core spreadsheet functionalities. It introduces a state-machine-based planning approach for robust multi-step interaction with spreadsheets. A dataset of 221 spreadsheet control tasks and an automated evaluation pipeline benchmark LLM software control performance, achieving 44.3% task completion for a single generation.",{"@graph":63,"@context":118},[64,80,101],{"@type":65,"itemListElement":66},"BreadcrumbList",[67,71,74,77],{"item":68,"name":69,"@type":70,"position":9},"https://docshare.wps.com","Home","ListItem",{"item":72,"name":10,"@type":70,"position":73},"https://docshare.wps.com/template/",2,{"item":75,"name":11,"@type":70,"position":76},"https://docshare.wps.com/template/presentations/",3,{"item":78,"name":59,"@type":70,"position":79},"https://docshare.wps.com/template/sheetcopilot-bringing-software-productivity-to-the-next-level-through-large-language-models/242153/",4,{"url":78,"name":59,"@type":81,"image":82,"author":87,"headline":59,"publisher":90,"fileFormat":93,"inLanguage":57,"description":61,"dateModified":94,"datePublished":95,"encodingFormat":93,"isAccessibleForFree":96,"interactionStatistic":97},"DigitalDocument",{"url":83,"@type":84,"width":85,"height":86},"https://docshare.wps.com/thumbnails/sheetcopilot-bringing-software-productivity-to-the-next-level-through-large-language-models/242153.png","ImageObject",442,249,{"name":88,"@type":89},"วิน","Person",{"url":68,"name":91,"@type":92},"DocShare","Organization","application/pdf","2026-10-11","2026-09-12",true,{"@type":98,"interactionType":99,"userInteractionCount":76},"InteractionCounter",{"@type":100},"ViewAction",{"@type":102,"mainEntity":103},"FAQPage",[104,110,114],{"name":105,"@type":106,"acceptedAnswer":107},"What problem does SheetCopilot target?","Question",{"text":108,"@type":109},"It targets the difficulty most end users face when trying to automate repetitive and error-prone spreadsheet tasks without programming skills, by enabling spreadsheet control through natural language.","Answer",{"name":111,"@type":106,"acceptedAnswer":112},"How does SheetCopilot interact with spreadsheets?",{"text":113,"@type":109},"It uses atomic actions to abstract spreadsheet functionalities and a state-machine-based task planning framework that decomposes natural-language requests into step-by-step plans, then issues executable commands.",{"name":115,"@type":106,"acceptedAnswer":116},"How is model performance evaluated in the paper?",{"text":117,"@type":109},"The work curates a dataset of 221 spreadsheet control tasks and provides a fully automated evaluation pipeline that benchmarks how accurately models comprehend requests, plan, and perform spreadsheet operations.","https://schema.org",{"og:url":78,"og:type":120,"og:title":59,"og:site_name":91,"og:description":61},"article",{"robots":122,"canonical":78},"index,follow",{"doc_id":124,"site_id":56},242153,1790829118,{"code":4,"msg":5,"data":127},{"doc_id":124,"user_id":128,"nickname":88,"user_avatar":129,"doc_module":9,"category_id":8,"category_name":11,"doc_title":59,"doc_description":61,"doc_content":130,"file_id":131,"file_url":132,"file_type":133,"file_size":134,"view_count":30,"is_deleted":4,"is_public":9,"is_downloadable":9,"audit_status":9,"page_count":135,"language":136,"language_code":57,"site_id":56,"html_lang":57,"table_of_contents":137,"faqs":138,"seo_title":139,"seo_description":61,"update_tm":140,"read_time":15},2336475104736,"https://ap-avatar.wpscdn.com/avatar/22000c4c5e0e5b17e70?x-image-process=image/resize,m_fixed,w_180,h_180&k=1786591360781797222","SheetCopilot: Bringing Software Productivity to the Next Level through Large Language Models  \nHongxin Li∗1,2, Jingran Su∗3,4, Yuntao Chen†3, Qing Li†4, and Zhaoxiang Zhang†1,2,3,5  \n1 School of Artificial Intelligence, University of Chinese Academy of Sciences (UCAS)  \n2 State Key Laboratory of Multimodal Artificial Intelligence Systems (MAIS), Institute of  \nAutomation, Chinese Academy of Sciences  \n3 Center for Artificial Intelligence and Robotics, HKISI, CAS  \n4The Hong Kong Polytechnic University  \n5 Shanghai Artificial Intelligence Laboratory  \nAbstract  \nComputer end users have spent billions of hours completing daily tasks like tabular data processing and project timeline scheduling. Most of these tasks are repetitive and error-prone, yet most end users lack the skill to automate these burdensome works. With the advent of large language models (LLMs), directing software with natural language user requests become a reachable goal. In this work, we propose a SheetCopilot agent that takes natural language task and control spreadsheet to fulfill the requirements. We propose a set of atomic actions as an abstraction of spreadsheet software functionalities. We further design a state machine-based task planning framework for LLMs to robustly interact with spreadsheets. We curate a representative dataset containing 221 spreadsheet control tasks and establish a fully automated evaluation pipeline for rigorously benchmarking the ability of LLMs in software control tasks. Our SheetCopilot correctly completes 44.3% of tasks fora single generation, outperforming the strong code generation baseline by a wide margin. Our project page: [https://sheetcopilot.github.io/](https://sheetcopilot.github.io/) .  \n1 Introduction  \nThe ability to intuitively direct sophisticated software through natural language has long been an aspiration pursued across generations. With the emergence of Large Language Models (LLMs) that can comprehend and respond to human directives, this vision is within closer reach now than ever. LLMs augmented with tools [24, 25, 20, 26, 2, 27] and reasoning abilities [32, 18, 30, 31] have shown promising results in recent research.  \nWhile LLMs continue advancing rapidly, their ability to interoperate seamlessly with existing software tools remains under-explored and not fully understood. Enabling LLMs to harness the rich functionality of countless existing software tools could unlock nearly limitless potential [21] .  \nProgress in endowing LLMs with the ability to direct complex software systems has been hindered by the lack of both a standardized framework for model-application interaction and a comprehensive benchmark for evaluating their performance. Several challenges exist in designing a framework to facilitate interaction between LLMs and sophisticated software applications, including 1) Translating the complex internal state and vast functionality of applications into text forms comprehensible for  \n∗ Equal contribution. E-mails: [lihongxin2021@ia.ac.cn](lihongxin2021@ia.ac.cn), [jing-ran.su@connect.polyu.hk](jing-ran.su@connect.polyu.hk)  \n† Equally advising corresponding authors. E-mails: [zhaoxiang.zhang@ia.ac.cn](zhaoxiang.zhang@ia.ac.cn), [csqli@comp.polyu.edu.hk](csqli@comp.polyu.edu.hk),  \n[chenyuntao08@gmail.com](chenyuntao08@gmail.com)  \n37th Conference on Neural Information Processing Systems (NeurIPS 2023) .  \nmodels [32] . This requires determining how to systematically represent software interfaces and logic through natural language; 2) Enabling models to generate software commands and parameters accurately and safely [9, 2] . Mechanisms must exist for validating, debugging, and as needed, rejecting or revising model outputs to avoid undesirable operations or states; 3) Providing models with means of monitoring software state changes, exceptions, and errors during multi-step tasks so that these models can respond appropriately. Models are required to understand software feedback and diagnose issues, ","cbCairpg97lUUncj","https://ap.wps.com/l/cbCairpg97lUUncj","pdf",12465671,33,"English","# Abstract\n# Introduction\n## Background: Natural-language software control\n## Motivation: Limited software interoperability and missing benchmarks\n## Framework and contributions: Agent design, atomic actions, dataset, evaluation","[{\"question\":\"What problem does SheetCopilot target?\",\"answer\":\"It targets the difficulty most end users face when trying to automate repetitive and error-prone spreadsheet tasks without programming skills, by enabling spreadsheet control through natural language.\"},{\"question\":\"How does SheetCopilot interact with spreadsheets?\",\"answer\":\"It uses atomic actions to abstract spreadsheet functionalities and a state-machine-based task planning framework that decomposes natural-language requests into step-by-step plans, then issues executable commands.\"},{\"question\":\"How is model performance evaluated in the paper?\",\"answer\":\"The work curates a dataset of 221 spreadsheet control tasks and provides a fully automated evaluation pipeline that benchmarks how accurately models comprehend requests, plan, and perform spreadsheet operations.\"}]","SheetCopilot - Bringing Software Productivity to the Next Level through Large Language Models | PDF",1789187040]