{"id":19070,"date":"2023-02-17T11:36:13","date_gmt":"2023-02-17T11:36:13","guid":{"rendered":"https:\/\/timi.eu\/?p=19070"},"modified":"2023-02-17T13:13:11","modified_gmt":"2023-02-17T13:13:11","slug":"chatgpt-etudie-grace-a-anatella","status":"publish","type":"post","link":"https:\/\/timi.eu\/fr\/blog\/anatella\/chatgpt-etudie-grace-a-anatella\/","title":{"rendered":"ChatGPT \u00e9tudi\u00e9 gr\u00e2ce \u00e0 Anatella"},"content":{"rendered":"\n\n<div\n\t class=\"wpml-ls-statics-shortcode_actions wpml-ls wpml-ls-legacy-dropdown js-wpml-ls-legacy-dropdown\">\n\t<ul>\n\n\t\t<li tabindex=\"0\" class=\"wpml-ls-slot-shortcode_actions wpml-ls-item wpml-ls-item-fr wpml-ls-current-language wpml-ls-item-legacy-dropdown\">\n\t\t\t<a href=\"#\" class=\"js-wpml-ls-item-toggle wpml-ls-item-toggle\">\n                                                    <img\n            class=\"wpml-ls-flag\"\n            src=\"https:\/\/timi.eu\/wp-content\/plugins\/sitepress-multilingual-cms\/res\/flags\/fr.png\"\n            alt=\"\"\n            width=18\n            height=12\n    \/><span class=\"wpml-ls-native\">Fran\u00e7ais<\/span><\/a>\n\n\t\t\t<ul class=\"wpml-ls-sub-menu\">\n\t\t\t\t\n\t\t\t\t\t<li class=\"wpml-ls-slot-shortcode_actions wpml-ls-item wpml-ls-item-en wpml-ls-first-item\">\n\t\t\t\t\t\t<a href=\"https:\/\/timi.eu\" class=\"wpml-ls-link\">\n                                                                <img\n            class=\"wpml-ls-flag\"\n            src=\"https:\/\/timi.eu\/wp-content\/plugins\/sitepress-multilingual-cms\/res\/flags\/en.png\"\n            alt=\"\"\n            width=18\n            height=12\n    \/><span class=\"wpml-ls-display\">Anglais<\/span><\/a>\n\t\t\t\t\t<\/li>\n\n\t\t\t\t\n\t\t\t\t\t<li class=\"wpml-ls-slot-shortcode_actions wpml-ls-item wpml-ls-item-es\">\n\t\t\t\t\t\t<a href=\"https:\/\/timi.eu\/es\/\" class=\"wpml-ls-link\">\n                                                                <img\n            class=\"wpml-ls-flag\"\n            src=\"https:\/\/timi.eu\/wp-content\/plugins\/sitepress-multilingual-cms\/res\/flags\/es.png\"\n            alt=\"\"\n            width=18\n            height=12\n    \/><span class=\"wpml-ls-display\">Espagnol<\/span><\/a>\n\t\t\t\t\t<\/li>\n\n\t\t\t\t\n\t\t\t\t\t<li class=\"wpml-ls-slot-shortcode_actions wpml-ls-item wpml-ls-item-nl\">\n\t\t\t\t\t\t<a href=\"https:\/\/timi.eu\/nl\/\" class=\"wpml-ls-link\">\n                                                                <img\n            class=\"wpml-ls-flag\"\n            src=\"https:\/\/timi.eu\/wp-content\/plugins\/sitepress-multilingual-cms\/res\/flags\/nl.png\"\n            alt=\"\"\n            width=18\n            height=12\n    \/><span class=\"wpml-ls-display\">N\u00e9erlandais<\/span><\/a>\n\t\t\t\t\t<\/li>\n\n\t\t\t\t\n\t\t\t\t\t<li class=\"wpml-ls-slot-shortcode_actions wpml-ls-item wpml-ls-item-ru wpml-ls-last-item\">\n\t\t\t\t\t\t<a href=\"https:\/\/timi.eu\/ru\/\" class=\"wpml-ls-link\">\n                                                                <img\n            class=\"wpml-ls-flag\"\n            src=\"https:\/\/timi.eu\/wp-content\/plugins\/sitepress-multilingual-cms\/res\/flags\/ru.png\"\n            alt=\"\"\n            width=18\n            height=12\n    \/><span class=\"wpml-ls-display\">Russe<\/span><\/a>\n\t\t\t\t\t<\/li>\n\n\t\t\t\t\t\t\t<\/ul>\n\n\t\t<\/li>\n\n\t<\/ul>\n<\/div>\n\n\n\n\n<h1 class=\"wp-block-heading\">ChatGPT \u00e9tudi\u00e9 gr\u00e2ce \u00e0 Anatella<\/h1>\n\n\n\n<h3 class=\"wp-block-heading\">Le <a href=\"https:\/\/www.intotheminds.com\/fr\/cabinet-etude-de-marche.html\" target=\"_blank\" rel=\"noreferrer noopener\">cabinet IntoTheMinds<\/a> vient de publier une nouvelle \u00e9tude r\u00e9alis\u00e9e gr\u00e2ce \u00e0 notre ETL : Anatella. Cette <a href=\"https:\/\/www.intotheminds.com\/blog\/chatgpt-similarite-avec-plan\/\" target=\"_blank\" rel=\"noreferrer noopener\">\u00e9tude<\/a> fait a lumi\u00e8re sur les textes g\u00e9n\u00e9r\u00e9s par ChatGPT et en particulier sur leur diversit\u00e9. La promesse marketing d&rsquo;OpenAi est en effet que chatGPT produise toujours des r\u00e9ponses diff\u00e9rentes.<\/h3>\n\n\n\n<p>Prise au pied de la lettre, cette affirmation est vraie. Encore fallait-il objectiver la similarit\u00e9 des r\u00e9ponses obtenues. L&rsquo;\u00e9tude publi\u00e9 le 16 F\u00e9vrier 2023 utilise un corpus de <strong>1000 textes<\/strong>. Vingt questions ont \u00e9t\u00e9 pos\u00e9es \u00e0 chatGPT et les r\u00e9ponses r\u00e9g\u00e9n\u00e9r\u00e9es 50 fois chacune.<\/p>\n\n\n\n<p>Les donn\u00e9es ont \u00e9t\u00e9 pr\u00e9par\u00e9es et analys\u00e9es gr\u00e2ce \u00e0 <a href=\"\/fr\/timi\/anatella\/\">Anatella<\/a>.<\/p>\n\n\n\n<p>Trois flux de donn\u00e9es ont \u00e9t\u00e9 utilis\u00e9s :<\/p>\n\n\n\n<ol>\n<li>un flux permet de calculer la similarit\u00e9 entre les it\u00e9rations pour une m\u00eame question<\/li>\n\n\n\n<li>un flux sp\u00e9cifique pour le calcul de la similarit\u00e9 entre chacun des 1000 textes<\/li>\n\n\n\n<li>un dernier flux se concentrant sur la similarit\u00e9 moyenne entre les r\u00e9ponses \u00e0 des questions diff\u00e9rentes<\/li>\n<\/ol>\n\n\n\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" loading=\"lazy\" width=\"1024\" height=\"422\" src=\"https:\/\/timi.eu\/wp-content\/uploads\/2023\/02\/pipeline-etl-anatella-chatGPT-experment-1-1024x422.png\" alt=\"\" class=\"wp-image-19071\" srcset=\"https:\/\/timi.eu\/wp-content\/uploads\/2023\/02\/pipeline-etl-anatella-chatGPT-experment-1-1024x422.png 1024w, https:\/\/timi.eu\/wp-content\/uploads\/2023\/02\/pipeline-etl-anatella-chatGPT-experment-1-300x124.png 300w, https:\/\/timi.eu\/wp-content\/uploads\/2023\/02\/pipeline-etl-anatella-chatGPT-experment-1-768x316.png 768w, https:\/\/timi.eu\/wp-content\/uploads\/2023\/02\/pipeline-etl-anatella-chatGPT-experment-1-1536x632.png 1536w, https:\/\/timi.eu\/wp-content\/uploads\/2023\/02\/pipeline-etl-anatella-chatGPT-experment-1-600x247.png 600w, https:\/\/timi.eu\/wp-content\/uploads\/2023\/02\/pipeline-etl-anatella-chatGPT-experment-1.png 1795w\" sizes=\"(max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<p><br>Les r\u00e9sultats sont riches :<\/p>\n\n\n\n<ul>\n<li>pour une m\u00eame question, la similarit\u00e9 entre les diff\u00e9rentes it\u00e9rations est en moyenne de <strong>75,3%<\/strong><\/li>\n\n\n\n<li>entre des questions diff\u00e9rentes, la similarit\u00e9 des r\u00e9ponses fournies est de <strong>60,9%<\/strong><\/li>\n\n\n\n<li>la similarit\u00e9 minimum mesur\u00e9e est de <strong>40%<\/strong><\/li>\n\n\n\n<li>la longueur des r\u00e9ponses de chatGPT varie \u00e9galement largement. Les 1000 textes produits montrent des variations maximales de <strong>+176% et -70%<\/strong>.<\/li>\n<\/ul>\n\n\n\n<p>Dans son rapport, IntoTheMinds met en avant la <strong>flexibilit\u00e9<\/strong> d&rsquo;Anatella, la <strong>rapidit\u00e9 d&rsquo;ex\u00e9cution<\/strong> (1 million de calculs de similarit\u00e9 en 44 secondes) ainsi que la <strong>richesse des outils<\/strong> fournies. Si les visualisations ont \u00e9t\u00e9 r\u00e9alis\u00e9es sous Tableau, l&rsquo;auteur de l&rsquo;\u00e9tude (Pierre-Nicolas Schwab) rel\u00e8ve que la transformation pr\u00e9alable des donn\u00e9es dans Anatella a \u00e9t\u00e9 indispensable pour contourner les limitations de Tableau.<\/p>\n","protected":false},"excerpt":{"rendered":"La promesse d&rsquo;OpenAi est que chatGPT produise toujours des r\u00e9ponses diff\u00e9rentes. Une nouvelle \u00e9tude r\u00e9alis\u00e9e gr\u00e2ce \u00e0 Anatella fait la lumi\u00e8re sur les textes g\u00e9n\u00e9r\u00e9s par ChatGPT et leur diversit\u00e9.","protected":false},"author":1,"featured_media":19076,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[248,3,55,184],"tags":[1001,1000,1003,309,269,999],"_links":{"self":[{"href":"https:\/\/timi.eu\/fr\/wp-json\/wp\/v2\/posts\/19070"}],"collection":[{"href":"https:\/\/timi.eu\/fr\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/timi.eu\/fr\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/timi.eu\/fr\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/timi.eu\/fr\/wp-json\/wp\/v2\/comments?post=19070"}],"version-history":[{"count":6,"href":"https:\/\/timi.eu\/fr\/wp-json\/wp\/v2\/posts\/19070\/revisions"}],"predecessor-version":[{"id":19172,"href":"https:\/\/timi.eu\/fr\/wp-json\/wp\/v2\/posts\/19070\/revisions\/19172"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/timi.eu\/fr\/wp-json\/wp\/v2\/media\/19076"}],"wp:attachment":[{"href":"https:\/\/timi.eu\/fr\/wp-json\/wp\/v2\/media?parent=19070"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/timi.eu\/fr\/wp-json\/wp\/v2\/categories?post=19070"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/timi.eu\/fr\/wp-json\/wp\/v2\/tags?post=19070"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}