{"id":1347,"date":"2025-12-14T16:01:15","date_gmt":"2025-12-14T15:01:15","guid":{"rendered":"https:\/\/www.ideas.edu.pl\/?post_type=publikacje&#038;p=1347"},"modified":"2025-12-14T16:01:16","modified_gmt":"2025-12-14T15:01:16","slug":"wait-thats-not-an-option-llms-robustness-with-incorrect-multiple-choice-options","status":"publish","type":"publikacje","link":"https:\/\/www.ideas.edu.pl\/en\/publikacje\/wait-thats-not-an-option-llms-robustness-with-incorrect-multiple-choice-options\/","title":{"rendered":"Wait, that's not an option: LLMs Robustness with Incorrect Multiple-Choice Options"},"content":{"rendered":"<p class=\"wp-block-paragraph\">This work introduces a novel framework for evaluating LLMs&#8217; capacity to balance instruction-following with critical reasoning when presented with multiple-choice questions containing no valid answers. Through systematic evaluation across arithmetic, domain-specific knowledge, and high-stakes medical decision tasks, we demonstrate that post-training aligned models often default to selecting invalid options, while base models exhibit improved refusal capabilities that scale with model size. Our analysis reveals that alignment techniques, though intended to enhance helpfulness, can inadvertently impair models&#8217; reflective judgment&#8211;the ability to override default behaviors when faced with invalid options. We additionally conduct a parallel human study showing similar instruction-following biases, with implications for how these biases may propagate through human feedback datasets used in alignment. We provide extensive ablation studies examining the impact of model size, training techniques, and prompt engineering. Our findings highlight fundamental tensions between alignment optimization and preservation of critical reasoning capabilities, with important implications for developing more robust AI systems for real-world deployment.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Autorzy: Gracjan G\u00f3ral, Emilia Wi\u015bnios, Piotr Sankowski, Pawe\u0142 Budzianowski<\/p>\n\n\n\n<div style=\"height:64px\" aria-hidden=\"true\" class=\"wp-block-spacer\"><\/div>\n\n\n\n<div class=\"wp-block-group is-layout-grid wp-container-core-group-is-layout-5ab56a3c wp-block-group-is-layout-grid\"><\/div>\n\n\n\n<div class=\"wp-block-group is-layout-grid wp-container-core-group-is-layout-5ab56a3c wp-block-group-is-layout-grid\"><\/div>","protected":false},"template":"","nazwa-konferencji":[28],"rodzaj-publikacji":[13],"rok-publikacji":[14],"class_list":["post-1347","publikacje","type-publikacje","status-publish","hentry","nazwa-konferencji-konferencja-acl","rodzaj-publikacji-artykul-konferencyjny","rok-publikacji-14"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v28.5 - https:\/\/yoast.com\/product\/yoast-seo-wordpress\/ -->\n<title>Wait, that&#039;s not an option: LLMs Robustness with Incorrect Multiple-Choice Options &#8226; IDEAS Instytut Badawczy<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.ideas.edu.pl\/en\/publikacje\/wait-thats-not-an-option-llms-robustness-with-incorrect-multiple-choice-options\/\" \/>\n<meta property=\"og:locale\" content=\"en_US\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Wait, that&#039;s not an option: LLMs Robustness with Incorrect Multiple-Choice Options &#8226; IDEAS Instytut Badawczy\" \/>\n<meta property=\"og:description\" content=\"This work introduces a novel framework for evaluating LLMs&#8217; capacity to balance instruction-following with critical reasoning when presented with multiple-choice questions containing no valid answers. Through systematic evaluation across arithmetic, domain-specific knowledge, and high-stakes medical decision tasks, we demonstrate that post-training aligned models often default to selecting invalid options, while base models exhibit improved refusal [&hellip;]\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.ideas.edu.pl\/en\/publikacje\/wait-thats-not-an-option-llms-robustness-with-incorrect-multiple-choice-options\/\" \/>\n<meta property=\"og:site_name\" content=\"IDEAS Instytut Badawczy\" \/>\n<meta property=\"article:modified_time\" content=\"2025-12-14T15:01:16+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/www.ideas.edu.pl\/wp-content\/uploads\/feature-image-home.webp\" \/>\n\t<meta property=\"og:image:width\" content=\"1800\" \/>\n\t<meta property=\"og:image:height\" content=\"945\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/webp\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"Est. reading time\" \/>\n\t<meta name=\"twitter:data1\" content=\"1 minute\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/www.ideas.edu.pl\\\/publikacje\\\/wait-thats-not-an-option-llms-robustness-with-incorrect-multiple-choice-options\\\/\",\"url\":\"https:\\\/\\\/www.ideas.edu.pl\\\/publikacje\\\/wait-thats-not-an-option-llms-robustness-with-incorrect-multiple-choice-options\\\/\",\"name\":\"Wait, that's not an option: LLMs Robustness with Incorrect Multiple-Choice Options &#8226; IDEAS Instytut Badawczy\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/www.ideas.edu.pl\\\/#website\"},\"datePublished\":\"2025-12-14T15:01:15+00:00\",\"dateModified\":\"2025-12-14T15:01:16+00:00\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/www.ideas.edu.pl\\\/publikacje\\\/wait-thats-not-an-option-llms-robustness-with-incorrect-multiple-choice-options\\\/#breadcrumb\"},\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/www.ideas.edu.pl\\\/publikacje\\\/wait-thats-not-an-option-llms-robustness-with-incorrect-multiple-choice-options\\\/\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/www.ideas.edu.pl\\\/publikacje\\\/wait-thats-not-an-option-llms-robustness-with-incorrect-multiple-choice-options\\\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Strona g\u0142\u00f3wna\",\"item\":\"https:\\\/\\\/www.ideas.edu.pl\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Publikacje\",\"item\":\"https:\\\/\\\/www.ideas.edu.pl\\\/publikacje\\\/\"},{\"@type\":\"ListItem\",\"position\":3,\"name\":\"Wait, that&#8217;s not an option: LLMs Robustness with Incorrect Multiple-Choice Options\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/www.ideas.edu.pl\\\/#website\",\"url\":\"https:\\\/\\\/www.ideas.edu.pl\\\/\",\"name\":\"IDEAS Instytut Badawczy\",\"description\":\"Pa\u0144stwowa jednostka badawczo-naukowa\",\"publisher\":{\"@id\":\"https:\\\/\\\/www.ideas.edu.pl\\\/#organization\"},\"alternateName\":\"IDEAS\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/www.ideas.edu.pl\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"en-US\"},{\"@type\":\"Organization\",\"@id\":\"https:\\\/\\\/www.ideas.edu.pl\\\/#organization\",\"name\":\"IDEAS Instytut Badawczy\",\"url\":\"https:\\\/\\\/www.ideas.edu.pl\\\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/www.ideas.edu.pl\\\/#\\\/schema\\\/logo\\\/image\\\/\",\"url\":\"https:\\\/\\\/www.ideas.edu.pl\\\/wp-content\\\/uploads\\\/e90241d6e73025d0d829abc28d67cb84.svg\",\"contentUrl\":\"https:\\\/\\\/www.ideas.edu.pl\\\/wp-content\\\/uploads\\\/e90241d6e73025d0d829abc28d67cb84.svg\",\"width\":152,\"height\":43,\"caption\":\"IDEAS Instytut Badawczy\"},\"image\":{\"@id\":\"https:\\\/\\\/www.ideas.edu.pl\\\/#\\\/schema\\\/logo\\\/image\\\/\"}}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"Wait, that's not an option: LLMs Robustness with Incorrect Multiple-Choice Options &#8226; IDEAS Instytut Badawczy","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.ideas.edu.pl\/en\/publikacje\/wait-thats-not-an-option-llms-robustness-with-incorrect-multiple-choice-options\/","og_locale":"en_US","og_type":"article","og_title":"Wait, that's not an option: LLMs Robustness with Incorrect Multiple-Choice Options &#8226; IDEAS Instytut Badawczy","og_description":"This work introduces a novel framework for evaluating LLMs&#8217; capacity to balance instruction-following with critical reasoning when presented with multiple-choice questions containing no valid answers. Through systematic evaluation across arithmetic, domain-specific knowledge, and high-stakes medical decision tasks, we demonstrate that post-training aligned models often default to selecting invalid options, while base models exhibit improved refusal [&hellip;]","og_url":"https:\/\/www.ideas.edu.pl\/en\/publikacje\/wait-thats-not-an-option-llms-robustness-with-incorrect-multiple-choice-options\/","og_site_name":"IDEAS Instytut Badawczy","article_modified_time":"2025-12-14T15:01:16+00:00","og_image":[{"width":1800,"height":945,"url":"https:\/\/www.ideas.edu.pl\/wp-content\/uploads\/feature-image-home.webp","type":"image\/webp"}],"twitter_card":"summary_large_image","twitter_misc":{"Est. reading time":"1 minute"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.ideas.edu.pl\/publikacje\/wait-thats-not-an-option-llms-robustness-with-incorrect-multiple-choice-options\/","url":"https:\/\/www.ideas.edu.pl\/publikacje\/wait-thats-not-an-option-llms-robustness-with-incorrect-multiple-choice-options\/","name":"Wait, that's not an option: LLMs Robustness with Incorrect Multiple-Choice Options &#8226; IDEAS Instytut Badawczy","isPartOf":{"@id":"https:\/\/www.ideas.edu.pl\/#website"},"datePublished":"2025-12-14T15:01:15+00:00","dateModified":"2025-12-14T15:01:16+00:00","breadcrumb":{"@id":"https:\/\/www.ideas.edu.pl\/publikacje\/wait-thats-not-an-option-llms-robustness-with-incorrect-multiple-choice-options\/#breadcrumb"},"inLanguage":"en-US","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.ideas.edu.pl\/publikacje\/wait-thats-not-an-option-llms-robustness-with-incorrect-multiple-choice-options\/"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.ideas.edu.pl\/publikacje\/wait-thats-not-an-option-llms-robustness-with-incorrect-multiple-choice-options\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Strona g\u0142\u00f3wna","item":"https:\/\/www.ideas.edu.pl\/"},{"@type":"ListItem","position":2,"name":"Publikacje","item":"https:\/\/www.ideas.edu.pl\/publikacje\/"},{"@type":"ListItem","position":3,"name":"Wait, that&#8217;s not an option: LLMs Robustness with Incorrect Multiple-Choice Options"}]},{"@type":"WebSite","@id":"https:\/\/www.ideas.edu.pl\/#website","url":"https:\/\/www.ideas.edu.pl\/","name":"IDEAS Research Institute","description":"State research and scientific unit","publisher":{"@id":"https:\/\/www.ideas.edu.pl\/#organization"},"alternateName":"IDEAS","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.ideas.edu.pl\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"en-US"},{"@type":"Organization","@id":"https:\/\/www.ideas.edu.pl\/#organization","name":"IDEAS Research Institute","url":"https:\/\/www.ideas.edu.pl\/","logo":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/www.ideas.edu.pl\/#\/schema\/logo\/image\/","url":"https:\/\/www.ideas.edu.pl\/wp-content\/uploads\/e90241d6e73025d0d829abc28d67cb84.svg","contentUrl":"https:\/\/www.ideas.edu.pl\/wp-content\/uploads\/e90241d6e73025d0d829abc28d67cb84.svg","width":152,"height":43,"caption":"IDEAS Instytut Badawczy"},"image":{"@id":"https:\/\/www.ideas.edu.pl\/#\/schema\/logo\/image\/"}}]}},"_links":{"self":[{"href":"https:\/\/www.ideas.edu.pl\/en\/wp-json\/wp\/v2\/publikacje\/1347","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.ideas.edu.pl\/en\/wp-json\/wp\/v2\/publikacje"}],"about":[{"href":"https:\/\/www.ideas.edu.pl\/en\/wp-json\/wp\/v2\/types\/publikacje"}],"wp:attachment":[{"href":"https:\/\/www.ideas.edu.pl\/en\/wp-json\/wp\/v2\/media?parent=1347"}],"wp:term":[{"taxonomy":"nazwa-konferencji","embeddable":true,"href":"https:\/\/www.ideas.edu.pl\/en\/wp-json\/wp\/v2\/nazwa-konferencji?post=1347"},{"taxonomy":"rodzaj-publikacji","embeddable":true,"href":"https:\/\/www.ideas.edu.pl\/en\/wp-json\/wp\/v2\/rodzaj-publikacji?post=1347"},{"taxonomy":"rok-publikacji","embeddable":true,"href":"https:\/\/www.ideas.edu.pl\/en\/wp-json\/wp\/v2\/rok-publikacji?post=1347"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}