{"id":1031829,"date":"2024-05-07T07:43:51","date_gmt":"2024-05-07T14:43:51","guid":{"rendered":"https:\/\/www.microsoft.com\/en-us\/research\/?post_type=msr-research-item&p=1031829"},"modified":"2024-05-07T07:43:51","modified_gmt":"2024-05-07T14:43:51","slug":"large-language-models-cannot-explain-themselves","status":"publish","type":"msr-research-item","link":"https:\/\/www.microsoft.com\/en-us\/research\/publication\/large-language-models-cannot-explain-themselves\/","title":{"rendered":"Large Language Models Cannot Explain Themselves"},"content":{"rendered":"

Large language models can be prompted to produce text. They can also be prompted to produce “explanations” of their output. But these are not really explanations, because they do not accurately reflect the mechanical process underlying the prediction. The illusion that they reflect the reasoning process can result in significant harms. These “explanations” can be valuable, but for promoting critical thinking rather than for understanding the model. I propose a recontextualisation of these “explanations”, using the term “exoplanations” to draw attention to their exogenous nature. I discuss some implications for design and technology, such as the inclusion of appropriate guardrails and responses when models are prompted to generate explanations.<\/p>\n","protected":false},"excerpt":{"rendered":"

Large language models can be prompted to produce text. They can also be prompted to produce “explanations” of their output. But these are not really explanations, because they do not accurately reflect the mechanical process underlying the prediction. The illusion that they reflect the reasoning process can result in significant harms. These “explanations” can be […]<\/p>\n","protected":false},"featured_media":0,"template":"","meta":{"msr-url-field":"","msr-podcast-episode":"","msrModifiedDate":"","msrModifiedDateEnabled":false,"ep_exclude_from_search":false,"_classifai_error":"","footnotes":""},"msr-content-type":[3],"msr-research-highlight":[],"research-area":[13556,13545,13554,13559],"msr-publication-type":[193716],"msr-product-type":[],"msr-focus-area":[],"msr-platform":[],"msr-download-source":[],"msr-locale":[268875],"msr-post-option":[],"msr-field-of-study":[246694,250138,248485,268089],"msr-conference":[260644],"msr-journal":[],"msr-impact-theme":[],"msr-pillar":[],"class_list":["post-1031829","msr-research-item","type-msr-research-item","status-publish","hentry","msr-research-area-artificial-intelligence","msr-research-area-human-language-technologies","msr-research-area-human-computer-interaction","msr-research-area-social-sciences","msr-locale-en_us","msr-field-of-study-artificial-intelligence","msr-field-of-study-generative-adversarial-network","msr-field-of-study-human-computer-interaction","msr-field-of-study-large-language-models"],"msr_publishername":"","msr_edition":"","msr_affiliation":"","msr_published_date":"2024-5-1","msr_host":"","msr_duration":"","msr_version":"","msr_speaker":"","msr_other_contributors":"","msr_booktitle":"","msr_pages_string":"","msr_chapter":"","msr_isbn":"","msr_journal":"","msr_volume":"","msr_number":"","msr_editors":"","msr_series":"","msr_issue":"","msr_organization":"","msr_how_published":"","msr_notes":"","msr_highlight_text":"","msr_release_tracker_id":"","msr_original_fields_of_study":"","msr_download_urls":"","msr_external_url":"","msr_secondary_video_url":"","msr_longbiography":"","msr_microsoftintellectualproperty":1,"msr_main_download":"","msr_publicationurl":"","msr_doi":"","msr_publication_uploader":[{"type":"file","viewUrl":"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2024\/05\/sarkar_2024_llms_cannot_explain.pdf","id":"1031841","title":"sarkar_2024_llms_cannot_explain","label_id":"243132","label":0}],"msr_related_uploader":"","msr_attachments":[{"id":1031841,"url":"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2024\/05\/sarkar_2024_llms_cannot_explain.pdf"}],"msr-author-ordering":[{"type":"user_nicename","value":"Advait Sarkar","user_id":37146,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Advait Sarkar"}],"msr_impact_theme":[],"msr_research_lab":[199561],"msr_event":[],"msr_group":[],"msr_project":[1053711,511097],"publication":[],"video":[],"download":[],"msr_publication_type":"inproceedings","related_content":{"projects":[{"ID":1053711,"post_title":"Tools for Thought","post_name":"t4t","post_type":"msr-project","post_date":"2024-07-03 05:21:39","post_modified":"2024-10-15 03:04:24","post_status":"publish","permalink":"https:\/\/www.microsoft.com\/en-us\/research\/project\/t4t\/","post_excerpt":"Better thinking through AI Much AI research focuses on solving specific tasks for people - generating content or automating processes. While such systems may be powerful, there are risks that this approach may bring to human cognition. They may impact the way people think and therefore learn, build skills, and grow expertise for example. The Tools for Thought (T4T) team aims to put human cognition at the heart of our systems. Our goal is to…","_links":{"self":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-project\/1053711"}]}},{"ID":511097,"post_title":"Calc Intelligence","post_name":"calc-intelligence","post_type":"msr-project","post_date":"2020-02-17 06:40:29","post_modified":"2023-11-27 06:33:39","post_status":"publish","permalink":"https:\/\/www.microsoft.com\/en-us\/research\/project\/calc-intelligence\/","post_excerpt":"By Calc Intelligence, we mean the research goal of bringing intelligence to end-user programming, and in particular to spreadsheets.","_links":{"self":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-project\/511097"}]}}]},"_links":{"self":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/1031829"}],"collection":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item"}],"about":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/types\/msr-research-item"}],"version-history":[{"count":1,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/1031829\/revisions"}],"predecessor-version":[{"id":1031844,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/1031829\/revisions\/1031844"}],"wp:attachment":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media?parent=1031829"}],"wp:term":[{"taxonomy":"msr-content-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-content-type?post=1031829"},{"taxonomy":"msr-research-highlight","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-highlight?post=1031829"},{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=1031829"},{"taxonomy":"msr-publication-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-publication-type?post=1031829"},{"taxonomy":"msr-product-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-product-type?post=1031829"},{"taxonomy":"msr-focus-area","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-focus-area?post=1031829"},{"taxonomy":"msr-platform","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-platform?post=1031829"},{"taxonomy":"msr-download-source","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-download-source?post=1031829"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=1031829"},{"taxonomy":"msr-post-option","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-post-option?post=1031829"},{"taxonomy":"msr-field-of-study","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-field-of-study?post=1031829"},{"taxonomy":"msr-conference","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-conference?post=1031829"},{"taxonomy":"msr-journal","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-journal?post=1031829"},{"taxonomy":"msr-impact-theme","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-impact-theme?post=1031829"},{"taxonomy":"msr-pillar","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-pillar?post=1031829"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}