{"id":1142,"date":"2024-06-25T09:00:00","date_gmt":"2024-06-25T07:00:00","guid":{"rendered":"https:\/\/smalltalks.eu\/blog\/?p=1142"},"modified":"2026-10-06T11:24:06","modified_gmt":"2026-10-06T09:24:06","slug":"comment-interpreter-ce-qui-se-passe-dans-un-reseau-de-neurones-alexandre-tl","status":"publish","type":"post","link":"https:\/\/smalltalks.eu\/blog\/2024\/06\/25\/comment-interpreter-ce-qui-se-passe-dans-un-reseau-de-neurones-alexandre-tl\/","title":{"rendered":"Comment interpr\u00e9ter ce qui se passe dans un r\u00e9seau de neurones ? (Alexandre TL)"},"content":{"rendered":"\n<figure class=\"wp-block-embed is-type-video is-provider-youtube wp-block-embed-youtube wp-embed-aspect-16-9 wp-has-aspect-ratio\"><div class=\"wp-block-embed__wrapper\">\n<iframe loading=\"lazy\" title=\"Comment interpr\u00e9ter ce qui se passe dans un r\u00e9seau de neurones ?\" width=\"640\" height=\"360\" src=\"https:\/\/www.youtube.com\/embed\/n4EnafoZ38Q?feature=oembed\" frameborder=\"0\" allow=\"accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share\" referrerpolicy=\"strict-origin-when-cross-origin\" allowfullscreen><\/iframe>\n<\/div><\/figure>\n\n\n\n<div style=\"height:30px\" aria-hidden=\"true\" class=\"wp-block-spacer\"><\/div>\n\n\n\n<p class=\"wp-block-paragraph\">On le dit souvent : les r\u00e9seaux de neurones sont des bo\u00eetes noires. Des mod\u00e8les comme ChatGPT ou Claude, avec des centaines de milliards de param\u00e8tres, seraient donc impossibles \u00e0 interpr\u00e9ter ? Pas tout \u00e0 fait.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Alexandre TL pr\u00e9sente une technique qui permet d&#8217;identifier des concepts r\u00e9els (par exemple la tour Eiffel, pour un mod\u00e8le de texte) dans les calculs du r\u00e9seau, puis de s&#8217;en servir pour contr\u00f4ler le mod\u00e8le de fa\u00e7on pr\u00e9visible. La vid\u00e9o s&#8217;appuie sur deux articles d&#8217;Anthropic, <em>Towards Monosemanticity<\/em> (2023) et <em>Scaling Monosemanticity<\/em> (2024).<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Voir aussi : <a href=\"https:\/\/smalltalks.eu\/blog\/?p=1138\">ce que les r\u00e9seaux de neurones apprennent vraiment<\/a>, <a href=\"https:\/\/smalltalks.eu\/blog\/?p=1143\">Platon et l&#8217;IA<\/a>.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><em>Vid\u00e9o d&#8217;Alexandre TL.<\/em><\/p>\n","protected":false},"excerpt":{"rendered":"<p>On le dit souvent : les r\u00e9seaux de neurones sont des bo\u00eetes noires. Des mod\u00e8les comme ChatGPT ou Claude, avec des centaines de milliards de param\u00e8tres, seraient donc impossibles \u00e0 interpr\u00e9ter ? Pas tout \u00e0 fait. Alexandre TL pr\u00e9sente une technique qui permet d&#8217;identifier des concepts r\u00e9els (par exemple la tour Eiffel, pour un mod\u00e8le de texte) dans les calculs du r\u00e9seau, puis de s&#8217;en servir pour contr\u00f4ler le mod\u00e8le de fa\u00e7on pr\u00e9visible. La vid\u00e9o s&#8217;appuie sur deux articles d&#8217;Anthropic,&#8230;<\/p>\n<p class=\"read-more\"><a class=\"btn btn-default\" href=\"https:\/\/smalltalks.eu\/blog\/2024\/06\/25\/comment-interpreter-ce-qui-se-passe-dans-un-reseau-de-neurones-alexandre-tl\/\"> Read More<span class=\"screen-reader-text\">  Read More<\/span><\/a><\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[110],"tags":[],"class_list":["post-1142","post","type-post","status-publish","format-standard","hentry","category-theme-intelligence-artificielle"],"_links":{"self":[{"href":"https:\/\/smalltalks.eu\/blog\/wp-json\/wp\/v2\/posts\/1142","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/smalltalks.eu\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/smalltalks.eu\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/smalltalks.eu\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/smalltalks.eu\/blog\/wp-json\/wp\/v2\/comments?post=1142"}],"version-history":[{"count":1,"href":"https:\/\/smalltalks.eu\/blog\/wp-json\/wp\/v2\/posts\/1142\/revisions"}],"predecessor-version":[{"id":1165,"href":"https:\/\/smalltalks.eu\/blog\/wp-json\/wp\/v2\/posts\/1142\/revisions\/1165"}],"wp:attachment":[{"href":"https:\/\/smalltalks.eu\/blog\/wp-json\/wp\/v2\/media?parent=1142"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/smalltalks.eu\/blog\/wp-json\/wp\/v2\/categories?post=1142"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/smalltalks.eu\/blog\/wp-json\/wp\/v2\/tags?post=1142"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}