[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-0-en-105":3,"doc-seo-273540-105":59,"doc-detail-273540-en":130},{"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":123,"head_meta":125,"extra_data":127,"updated_unix":129},105,"en","bayesprompt-human-readable-prompts-that-make-sense","BAYESPROMPT - Human Readable Prompts That Make Sense","","Reconstructing prompts that elicit a desired answer or behavior in a large language model is an open research problem. Existing optimization methods that minimize perplexity often produce pseudoprompts: token strings that steer the model yet remain unintelligible to humans. This work argues that the task is ill-posed and reframes prompt reconstruction as Bayesian posterior inference, enabling efficient sampling of prompts that remain low-perplexity and human readable. Experiments on real data show clear metric improvements over state of the art.",{"@graph":69,"@context":122},[70,84,105],{"@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/bayesprompt-human-readable-prompts-that-make-sense/273540/",{"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/bayesprompt-human-readable-prompts-that-make-sense/273540.png","ImageObject",300,407,{"name":92,"@type":93},"Maeve","Person",{"url":74,"name":95,"@type":96},"DocShare","Organization","application/pdf","2026-10-30","2026-09-15",true,{"@type":102,"interactionType":103,"userInteractionCount":52},"InteractionCounter",{"@type":104},"ViewAction",{"@type":106,"mainEntity":107},"FAQPage",[108,114,118],{"name":109,"@type":110,"acceptedAnswer":111},"Why do prompt optimization methods often produce pseudoprompts?","Question",{"text":112,"@type":113},"They minimize perplexity but ignore aspects that make prompt reconstruction ill-posed, leading to token sequences that steer the model without being human interpretable.","Answer",{"name":115,"@type":110,"acceptedAnswer":116},"How does Bayesian inference help reconstruct human-readable prompts?",{"text":117,"@type":113},"It reframes prompt optimization as Bayesian posterior inference over prompts, allowing efficient sampling of prompts that balance low perplexity with human readability.",{"name":119,"@type":110,"acceptedAnswer":120},"What method is proposed to sample prompts in the Bayesian framework?",{"text":121,"@type":113},"An efficient algorithm based on Markov-chain Monte Carlo is proposed to obtain prompts with strong performance and better interpretability.","https://schema.org",{"og:url":83,"og:type":124,"og:title":65,"og:site_name":95,"og:description":67},"article",{"robots":126,"canonical":83},"index,follow",{"doc_id":128,"site_id":62},273540,1789471963,{"code":4,"msg":5,"data":131},{"doc_id":128,"user_id":132,"nickname":92,"user_avatar":133,"doc_module":4,"category_id":39,"category_name":40,"doc_title":65,"doc_description":67,"doc_content":134,"file_id":135,"file_url":136,"file_type":137,"file_size":138,"view_count":52,"is_deleted":4,"is_public":8,"is_downloadable":8,"audit_status":8,"page_count":139,"language":140,"language_code":63,"site_id":62,"html_lang":63,"table_of_contents":141,"faqs":142,"seo_title":143,"seo_description":67,"update_tm":129,"read_time":36,"preview_page_count":4},5909877438554,"https://ap-avatar.wpscdn.com/avatar/5600025385ad2bf12a7?_k=1778553567797529272","arXiv : 2608 . 17866v 1 [cs .CL] 18 Aug 2026  \nBAYESPROMPT: HUMAN READABLE PROMPTS THAT MAKE SENSE  \nFranky Kevin Nando Tezoh, Ali Hussaini Umar, Alessandro Laio, & Guido Sanguinetti ∗ Department of Physics,  \nScuola Internazionale Superiore di Studi Avanzati (SISSA), Trieste, Italy,  \n{fnandote,aumar,laio,[gsanguin}@sissa.it](gsanguin}@sissa.it)  \nRiccardo Rende  \nCenter for Computational Quantum Physics, Flatiron Institute,  \n162 5th Avenue, New York, NY 10010,{ [rrende}@flatironinstitute.org](rrende}@flatironinstitute.org)  \nABSTRACT  \nReconstructing prompts that can elicit a desired answer or behaviour in an LLMis an open and important research topic. Optimisation methods which aim at minimising the perplexity of a given answer, however, consistently yield socalled pseudoprompts, unintelligible strings of tokens which can lack human interpretability. We argue that this is a consequence of the ill-posedness of the prompt optimisation task. By reframing the task as a Bayesian posterior inference over prompts, we propose an efficient algorithm to sample prompts which are both efficient (in terms of perplexity) and human readable. We compare our approach with state of the art alternatives showing on a real data set a marked improvement over a range of metrics.  \n1 INTRODUCTION  \nIn-context learning (ICL) (Dong et al., 2022) has recently become a major focus in machine learning. Researchers found that Large Language Models (LLMs) can solve diverse tasks like translating languages, analyzing sentiments, or solving math problems simply by conditioning on a few exemplars within a prompt. They do this without requiring any update to the model’s internal parameters. Much research has been dedicated to trying to explain what gives rise to the surprising in-context learning behaviour of LLMs (Rai et al., 2024; Singh et al., 2024; Vuli et al., 2020; Jin et al., 2024; Ju et al., 2024; Elhage et al., 2021; Olsson et al., 2022); regardless of its origin, the ICL phenomenon has also been immediately recognised for its practical importance.  \nA key ingredient for ICL is designing the correct prompt for a given task. Prompt optimisation approaches (Sinha et al., 2021) aim to discover high confidence (low perplexity) prompts to leverage ICL for practical purposes, using both continuous soft prompts (Lester et al., 2021; Liu et al., 2024) and discrete hard prompts (Wen et al., 2023) to construct more effective inputs.  \nMany studies have addressed the prompt-optimisation problem, revealing a striking gap between human intuition and model behavior: optimal prompts are often short non-sensical sequences of tokens that nonetheless reliably steer a model towards a target output. Such pseudoprompts were first identified through universal adversarial triggers (Wallace et al., 2019); this phenomenon has since been shown to generalize, as models conditioned on unnatural, human-illegible sequences can match or exceed the performance of natural prompts (Zou et al., 2023; Kervadec et al., 2023; Cherepanova & Zou, 2024), consistent with evidence that models may rely more on distributional statistics than on human syntactic structure (Sinha et al., 2021) . This suggests that the space of effective prompts is far larger, and far less human readable, than the space of prompts a person would ever write. Subsequent  \n∗ Correspondence should be [addressed to laio@sissa.it](addressed to laio@sissa.it), [gsangiun@sissa.it](gsangiun@sissa.it), [fnandote@sissa.it](fnandote@sissa.it).  \nstudies on Unnatural Language Processing broadened this observation, challenging the assumption that LLMs rely on human-centric linguistic structures: (Zou et al., 2023; Kervadec et al., 2023; Cherepanova & Zou, 2024) demonstrated that models prompted with unnatural sequences which are semantically and syntactically incoherent to humans still produce predictable, meaningful outputs.  \nContribution: In this paper, we aim to illustrate the statistical origins of pseudoprompts and to pr","cbCaidj4Hlg7uPx8","https://ap.wps.com/l/cbCaidj4Hlg7uPx8","pdf",975842,16,"English","# Abstract\n# 1 Introduction\n# 2 Related Work","[{\"question\":\"Why do prompt optimization methods often produce pseudoprompts?\",\"answer\":\"They minimize perplexity but ignore aspects that make prompt reconstruction ill-posed, leading to token sequences that steer the model without being human interpretable.\"},{\"question\":\"How does Bayesian inference help reconstruct human-readable prompts?\",\"answer\":\"It reframes prompt optimization as Bayesian posterior inference over prompts, allowing efficient sampling of prompts that balance low perplexity with human readability.\"},{\"question\":\"What method is proposed to sample prompts in the Bayesian framework?\",\"answer\":\"An efficient algorithm based on Markov-chain Monte Carlo is proposed to obtain prompts with strong performance and better interpretability.\"}]","BAYESPROMPT - Human Readable Prompts That Make Sense | PDF"]