{"id":10653,"date":"2024-03-12T10:07:58","date_gmt":"2024-03-12T10:07:58","guid":{"rendered":"https:\/\/dailyai.com\/?p=10653"},"modified":"2024-03-12T10:07:58","modified_gmt":"2024-03-12T10:07:58","slug":"wmdp-measures-and-reduces-llm-malicious-use-with-unlearning","status":"publish","type":"post","link":"https:\/\/dailyai.com\/fr\/2024\/03\/wmdp-measures-and-reduces-llm-malicious-use-with-unlearning\/","title":{"rendered":"WMDP mesure et r\u00e9duit l'utilisation malveillante du LLM avec d\u00e9sapprentissage"},"content":{"rendered":"

Les chercheurs ont publi\u00e9 un crit\u00e8re de r\u00e9f\u00e9rence pour mesurer si un LLM contient des connaissances potentiellement dangereuses et une nouvelle technique pour d\u00e9sapprendre les donn\u00e9es dangereuses.<\/strong><\/p>\n

La question de savoir si les mod\u00e8les d'intelligence artificielle pourraient aider des acteurs malveillants \u00e0 fabriquer une bombe, \u00e0 planifier un attentat, etc. a fait l'objet de nombreux d\u00e9bats. attaque de cybers\u00e9curit\u00e9<\/a>ou fabriquer une arme biologique<\/a>.<\/p>\n

L'\u00e9quipe de chercheurs de Scale AI, du Center for AI Safety et d'experts d'\u00e9tablissements d'enseignement de premier plan a publi\u00e9 une \u00e9tude comparative qui nous permet de mieux mesurer le degr\u00e9 de dangerosit\u00e9 d'un programme d'\u00e9ducation et de formation tout au long de la vie.<\/p>\n

Le test de r\u00e9f\u00e9rence sur les armes de destruction massive (WMDP) est un ensemble de 4 157 questions \u00e0 choix multiples portant sur les connaissances dangereuses en mati\u00e8re de bios\u00e9curit\u00e9, de cybers\u00e9curit\u00e9 et de s\u00e9curit\u00e9 chimique.<\/p>\n

Plus le score d'un LLM est \u00e9lev\u00e9, plus il pr\u00e9sente un danger en permettant \u00e0 une personne d'avoir des intentions criminelles. Un LLM ayant un score WMDP plus faible est moins susceptible de vous aider \u00e0 fabriquer une bombe ou \u00e0 cr\u00e9er un nouveau virus.<\/p>\n

La mani\u00e8re traditionnelle de rendre un LLM plus conforme est de refuser les demandes de donn\u00e9es qui pourraient permettre des actions malveillantes. Le jailbreaking ou peaufinage<\/a> un LLM align\u00e9 pourrait supprimer ces garde-fous et exposer des connaissances dangereuses dans l'ensemble de donn\u00e9es du mod\u00e8le.<\/p>\n

Si vous pouviez faire en sorte que le mod\u00e8le oublie ou d\u00e9sapprenne l'information incrimin\u00e9e, il n'y aurait aucun risque qu'il la d\u00e9livre par inadvertance en r\u00e9ponse \u00e0 une demande intelligente de la part de l'entreprise. d\u00e9verrouillage<\/a> technique.<\/p>\n

En leur document de recherche<\/a>Les chercheurs expliquent comment ils ont d\u00e9velopp\u00e9 un algorithme appel\u00e9 Contrastive Unlearn Tuning (CUT), une m\u00e9thode de r\u00e9glage fin permettant de d\u00e9sapprendre les connaissances dangereuses tout en conservant les informations b\u00e9nignes.<\/p>\n

La m\u00e9thode de r\u00e9glage fin CUT proc\u00e8de au d\u00e9sapprentissage de la machine en optimisant un \"terme d'oubli\" de sorte que le mod\u00e8le devienne moins expert sur les sujets dangereux. Elle optimise \u00e9galement un \"terme de r\u00e9tention\" afin que le mod\u00e8le fournisse des r\u00e9ponses utiles \u00e0 des demandes b\u00e9nignes.<\/p>\n

La nature \u00e0 double usage d'une grande partie des informations contenues dans les ensembles de donn\u00e9es de formation LLM fait qu'il est difficile de ne d\u00e9sapprendre que les mauvaises choses tout en conservant les informations utiles. En utilisant WMDP, les chercheurs ont pu construire des ensembles de donn\u00e9es \"oublier\" et \"conserver\" pour diriger leur technique de d\u00e9sapprentissage CUT.<\/p>\n

Les chercheurs ont utilis\u00e9 le WMDP pour mesurer la probabilit\u00e9 que le mod\u00e8le ZEPHYR-7B-BETA fournisse des informations dangereuses avant et apr\u00e8s le d\u00e9sapprentissage \u00e0 l'aide de la CUT. Leurs tests ont port\u00e9 sur la bios\u00e9curit\u00e9 et la cybers\u00e9curit\u00e9.<\/p>\n

Ils ont ensuite test\u00e9 le mod\u00e8le pour voir si ses performances g\u00e9n\u00e9rales avaient souffert du processus de d\u00e9sapprentissage.<\/p>\n

\"\"
Pr\u00e9cision du WMPD et performance du MMLU avant et apr\u00e8s le d\u00e9sapprentissage de la CUT. Source : arXiv<\/figcaption><\/figure>\n

Les r\u00e9sultats montrent que le processus de d\u00e9sapprentissage a consid\u00e9rablement r\u00e9duit la pr\u00e9cision des r\u00e9ponses aux demandes dangereuses avec seulement une r\u00e9duction marginale de la performance du mod\u00e8le sur le benchmark MMLU.<\/p>\n

Malheureusement, le CUT r\u00e9duit la pr\u00e9cision des r\u00e9ponses dans des domaines \u00e9troitement li\u00e9s tels que l'introduction \u00e0 la virologie et la s\u00e9curit\u00e9 informatique. Fournir une r\u00e9ponse utile \u00e0 la question \"Comment arr\u00eater une cyber-attaque ?\" mais pas \u00e0 la question \"Comment mener une cyber-attaque ?\" exige une plus grande pr\u00e9cision dans le processus de d\u00e9sapprentissage.<\/p>\n

Les chercheurs ont \u00e9galement constat\u00e9 qu'ils ne pouvaient pas \u00e9liminer avec pr\u00e9cision les connaissances chimiques dangereuses, car elles \u00e9taient trop \u00e9troitement li\u00e9es aux connaissances chimiques g\u00e9n\u00e9rales.<\/p>\n

En utilisant CUT, les fournisseurs de mod\u00e8les ferm\u00e9s comme le GPT-4 pourraient d\u00e9sapprendre les informations dangereuses, de sorte que m\u00eame s'ils sont soumis \u00e0 une mise au point malveillante ou \u00e0 un \"jailbreaking\", ils ne se souviennent d'aucune information dangereuse \u00e0 d\u00e9livrer.<\/p>\n

Il est possible de faire de m\u00eame avec les mod\u00e8les \u00e0 source ouverte, mais l'acc\u00e8s public \u00e0 leurs poids signifie qu'ils peuvent r\u00e9apprendre des donn\u00e9es dangereuses s'ils sont entra\u00een\u00e9s sur ces derni\u00e8res.<\/p>\n

Cette m\u00e9thode de d\u00e9sapprentissage des donn\u00e9es dangereuses par un mod\u00e8le d'IA n'est pas infaillible, en particulier pour les mod\u00e8les open-source, mais elle constitue un ajout solide aux mod\u00e8les d'IA actuels. alignement<\/a> des m\u00e9thodes.<\/p>","protected":false},"excerpt":{"rendered":"

Les chercheurs ont publi\u00e9 un crit\u00e8re permettant de mesurer si un LLM contient des connaissances potentiellement dangereuses, ainsi qu'une nouvelle technique pour d\u00e9sapprendre les donn\u00e9es dangereuses. La question de savoir si les mod\u00e8les d'IA pourraient aider de mauvais acteurs \u00e0 fabriquer une bombe, \u00e0 planifier une attaque de cybers\u00e9curit\u00e9 ou \u00e0 fabriquer une arme biologique a fait l'objet de nombreux d\u00e9bats. L'\u00e9quipe de chercheurs de Scale AI, du Center for AI Safety et d'experts d'\u00e9tablissements d'enseignement de premier plan a publi\u00e9 une analyse comparative qui nous permet de mieux mesurer le degr\u00e9 de dangerosit\u00e9 d'un mod\u00e8le d'apprentissage tout au long de la vie. Le r\u00e9f\u00e9rentiel Weapons of Mass Destruction Proxy (WMDP) est un ensemble de 4 157 questions \u00e0 choix multiples portant sur des connaissances dangereuses en mati\u00e8re de bios\u00e9curit\u00e9,<\/p>","protected":false},"author":6,"featured_media":10656,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[84],"tags":[339,118],"class_list":["post-10653","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-industry","tag-ai-safety","tag-llms"],"yoast_head":"\nWMDP measures and reduces LLM malicious use with unlearning | DailyAI<\/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:\/\/dailyai.com\/fr\/2024\/03\/wmdp-measures-and-reduces-llm-malicious-use-with-unlearning\/\" \/>\n<meta property=\"og:locale\" content=\"fr_FR\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"WMDP measures and reduces LLM malicious use with unlearning | DailyAI\" \/>\n<meta property=\"og:description\" content=\"Researchers released a benchmark to measure whether an LLM contains potentially hazardous knowledge and a novel technique for unlearning dangerous data. There has been much debate over whether AI models could help bad actors build a bomb, plan a cybersecurity attack, or build a bioweapon. The team of researchers from Scale AI, the Center for AI Safety, and experts from leading educational institutions, released a benchmark that gives us a better measure of just how dangerous a particular LLM is. The Weapons of Mass Destruction Proxy (WMDP) benchmark is a dataset of 4,157 multiple-choice questions surrounding hazardous knowledge in biosecurity,\" \/>\n<meta property=\"og:url\" content=\"https:\/\/dailyai.com\/fr\/2024\/03\/wmdp-measures-and-reduces-llm-malicious-use-with-unlearning\/\" \/>\n<meta property=\"og:site_name\" content=\"DailyAI\" \/>\n<meta property=\"article:published_time\" content=\"2024-03-12T10:07:58+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/dailyai.com\/wp-content\/uploads\/2024\/03\/AI-unlearning.jpg\" \/>\n\t<meta property=\"og:image:width\" content=\"1640\" \/>\n\t<meta property=\"og:image:height\" content=\"924\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/jpeg\" \/>\n<meta name=\"author\" content=\"Eugene van der Watt\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:creator\" content=\"@DailyAIOfficial\" \/>\n<meta name=\"twitter:site\" content=\"@DailyAIOfficial\" \/>\n<meta name=\"twitter:label1\" content=\"\u00c9crit par\" \/>\n\t<meta name=\"twitter:data1\" content=\"Eugene van der Watt\" \/>\n\t<meta name=\"twitter:label2\" content=\"Dur\u00e9e de lecture estim\u00e9e\" \/>\n\t<meta name=\"twitter:data2\" content=\"3 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"NewsArticle\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/03\\\/wmdp-measures-and-reduces-llm-malicious-use-with-unlearning\\\/#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/03\\\/wmdp-measures-and-reduces-llm-malicious-use-with-unlearning\\\/\"},\"author\":{\"name\":\"Eugene van der Watt\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/#\\\/schema\\\/person\\\/7ce525c6d0c79838b7cc7cde96993cfa\"},\"headline\":\"WMDP measures and reduces LLM malicious use with unlearning\",\"datePublished\":\"2024-03-12T10:07:58+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/03\\\/wmdp-measures-and-reduces-llm-malicious-use-with-unlearning\\\/\"},\"wordCount\":583,\"publisher\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/#organization\"},\"image\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/03\\\/wmdp-measures-and-reduces-llm-malicious-use-with-unlearning\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/dailyai.com\\\/wp-content\\\/uploads\\\/2024\\\/03\\\/AI-unlearning.jpg\",\"keywords\":[\"AI safety\",\"LLMS\"],\"articleSection\":[\"Industry\"],\"inLanguage\":\"fr-FR\"},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/03\\\/wmdp-measures-and-reduces-llm-malicious-use-with-unlearning\\\/\",\"url\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/03\\\/wmdp-measures-and-reduces-llm-malicious-use-with-unlearning\\\/\",\"name\":\"WMDP measures and reduces LLM malicious use with unlearning | DailyAI\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/03\\\/wmdp-measures-and-reduces-llm-malicious-use-with-unlearning\\\/#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/03\\\/wmdp-measures-and-reduces-llm-malicious-use-with-unlearning\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/dailyai.com\\\/wp-content\\\/uploads\\\/2024\\\/03\\\/AI-unlearning.jpg\",\"datePublished\":\"2024-03-12T10:07:58+00:00\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/03\\\/wmdp-measures-and-reduces-llm-malicious-use-with-unlearning\\\/#breadcrumb\"},\"inLanguage\":\"fr-FR\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/dailyai.com\\\/2024\\\/03\\\/wmdp-measures-and-reduces-llm-malicious-use-with-unlearning\\\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"fr-FR\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/03\\\/wmdp-measures-and-reduces-llm-malicious-use-with-unlearning\\\/#primaryimage\",\"url\":\"https:\\\/\\\/dailyai.com\\\/wp-content\\\/uploads\\\/2024\\\/03\\\/AI-unlearning.jpg\",\"contentUrl\":\"https:\\\/\\\/dailyai.com\\\/wp-content\\\/uploads\\\/2024\\\/03\\\/AI-unlearning.jpg\",\"width\":1640,\"height\":924},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/03\\\/wmdp-measures-and-reduces-llm-malicious-use-with-unlearning\\\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\\\/\\\/dailyai.com\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"WMDP measures and reduces LLM malicious use with unlearning\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/#website\",\"url\":\"https:\\\/\\\/dailyai.com\\\/\",\"name\":\"DailyAI\",\"description\":\"Your Daily Dose of AI News\",\"publisher\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/#organization\"},\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/dailyai.com\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"fr-FR\"},{\"@type\":\"Organization\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/#organization\",\"name\":\"DailyAI\",\"url\":\"https:\\\/\\\/dailyai.com\\\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"fr-FR\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/#\\\/schema\\\/logo\\\/image\\\/\",\"url\":\"https:\\\/\\\/dailyai.com\\\/wp-content\\\/uploads\\\/2023\\\/06\\\/Daily-Ai_TL_colour.png\",\"contentUrl\":\"https:\\\/\\\/dailyai.com\\\/wp-content\\\/uploads\\\/2023\\\/06\\\/Daily-Ai_TL_colour.png\",\"width\":4501,\"height\":934,\"caption\":\"DailyAI\"},\"image\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/#\\\/schema\\\/logo\\\/image\\\/\"},\"sameAs\":[\"https:\\\/\\\/x.com\\\/DailyAIOfficial\",\"https:\\\/\\\/www.linkedin.com\\\/company\\\/dailyaiofficial\\\/\",\"https:\\\/\\\/www.youtube.com\\\/@DailyAIOfficial\"]},{\"@type\":\"Person\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/#\\\/schema\\\/person\\\/7ce525c6d0c79838b7cc7cde96993cfa\",\"name\":\"Eugene van der Watt\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"fr-FR\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/wp-content\\\/uploads\\\/2023\\\/07\\\/Eugine_Profile_Picture-96x96.png\",\"url\":\"https:\\\/\\\/dailyai.com\\\/wp-content\\\/uploads\\\/2023\\\/07\\\/Eugine_Profile_Picture-96x96.png\",\"contentUrl\":\"https:\\\/\\\/dailyai.com\\\/wp-content\\\/uploads\\\/2023\\\/07\\\/Eugine_Profile_Picture-96x96.png\",\"caption\":\"Eugene van der Watt\"},\"description\":\"Eugene comes from an electronic engineering background and loves all things tech. When he takes a break from consuming AI news you'll find him at the snooker table.\",\"sameAs\":[\"www.linkedin.com\\\/in\\\/eugene-van-der-watt-16828119\"],\"url\":\"https:\\\/\\\/dailyai.com\\\/fr\\\/author\\\/eugene\\\/\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"WMDP mesure et r\u00e9duit l'utilisation malveillante de LLM avec d\u00e9sapprentissage | DailyAI","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:\/\/dailyai.com\/fr\/2024\/03\/wmdp-measures-and-reduces-llm-malicious-use-with-unlearning\/","og_locale":"fr_FR","og_type":"article","og_title":"WMDP measures and reduces LLM malicious use with unlearning | DailyAI","og_description":"Researchers released a benchmark to measure whether an LLM contains potentially hazardous knowledge and a novel technique for unlearning dangerous data. There has been much debate over whether AI models could help bad actors build a bomb, plan a cybersecurity attack, or build a bioweapon. The team of researchers from Scale AI, the Center for AI Safety, and experts from leading educational institutions, released a benchmark that gives us a better measure of just how dangerous a particular LLM is. The Weapons of Mass Destruction Proxy (WMDP) benchmark is a dataset of 4,157 multiple-choice questions surrounding hazardous knowledge in biosecurity,","og_url":"https:\/\/dailyai.com\/fr\/2024\/03\/wmdp-measures-and-reduces-llm-malicious-use-with-unlearning\/","og_site_name":"DailyAI","article_published_time":"2024-03-12T10:07:58+00:00","og_image":[{"width":1640,"height":924,"url":"https:\/\/dailyai.com\/wp-content\/uploads\/2024\/03\/AI-unlearning.jpg","type":"image\/jpeg"}],"author":"Eugene van der Watt","twitter_card":"summary_large_image","twitter_creator":"@DailyAIOfficial","twitter_site":"@DailyAIOfficial","twitter_misc":{"\u00c9crit par":"Eugene van der Watt","Dur\u00e9e de lecture estim\u00e9e":"3 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"NewsArticle","@id":"https:\/\/dailyai.com\/2024\/03\/wmdp-measures-and-reduces-llm-malicious-use-with-unlearning\/#article","isPartOf":{"@id":"https:\/\/dailyai.com\/2024\/03\/wmdp-measures-and-reduces-llm-malicious-use-with-unlearning\/"},"author":{"name":"Eugene van der Watt","@id":"https:\/\/dailyai.com\/#\/schema\/person\/7ce525c6d0c79838b7cc7cde96993cfa"},"headline":"WMDP measures and reduces LLM malicious use with unlearning","datePublished":"2024-03-12T10:07:58+00:00","mainEntityOfPage":{"@id":"https:\/\/dailyai.com\/2024\/03\/wmdp-measures-and-reduces-llm-malicious-use-with-unlearning\/"},"wordCount":583,"publisher":{"@id":"https:\/\/dailyai.com\/#organization"},"image":{"@id":"https:\/\/dailyai.com\/2024\/03\/wmdp-measures-and-reduces-llm-malicious-use-with-unlearning\/#primaryimage"},"thumbnailUrl":"https:\/\/dailyai.com\/wp-content\/uploads\/2024\/03\/AI-unlearning.jpg","keywords":["AI safety","LLMS"],"articleSection":["Industry"],"inLanguage":"fr-FR"},{"@type":"WebPage","@id":"https:\/\/dailyai.com\/2024\/03\/wmdp-measures-and-reduces-llm-malicious-use-with-unlearning\/","url":"https:\/\/dailyai.com\/2024\/03\/wmdp-measures-and-reduces-llm-malicious-use-with-unlearning\/","name":"WMDP mesure et r\u00e9duit l'utilisation malveillante de LLM avec d\u00e9sapprentissage | DailyAI","isPartOf":{"@id":"https:\/\/dailyai.com\/#website"},"primaryImageOfPage":{"@id":"https:\/\/dailyai.com\/2024\/03\/wmdp-measures-and-reduces-llm-malicious-use-with-unlearning\/#primaryimage"},"image":{"@id":"https:\/\/dailyai.com\/2024\/03\/wmdp-measures-and-reduces-llm-malicious-use-with-unlearning\/#primaryimage"},"thumbnailUrl":"https:\/\/dailyai.com\/wp-content\/uploads\/2024\/03\/AI-unlearning.jpg","datePublished":"2024-03-12T10:07:58+00:00","breadcrumb":{"@id":"https:\/\/dailyai.com\/2024\/03\/wmdp-measures-and-reduces-llm-malicious-use-with-unlearning\/#breadcrumb"},"inLanguage":"fr-FR","potentialAction":[{"@type":"ReadAction","target":["https:\/\/dailyai.com\/2024\/03\/wmdp-measures-and-reduces-llm-malicious-use-with-unlearning\/"]}]},{"@type":"ImageObject","inLanguage":"fr-FR","@id":"https:\/\/dailyai.com\/2024\/03\/wmdp-measures-and-reduces-llm-malicious-use-with-unlearning\/#primaryimage","url":"https:\/\/dailyai.com\/wp-content\/uploads\/2024\/03\/AI-unlearning.jpg","contentUrl":"https:\/\/dailyai.com\/wp-content\/uploads\/2024\/03\/AI-unlearning.jpg","width":1640,"height":924},{"@type":"BreadcrumbList","@id":"https:\/\/dailyai.com\/2024\/03\/wmdp-measures-and-reduces-llm-malicious-use-with-unlearning\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/dailyai.com\/"},{"@type":"ListItem","position":2,"name":"WMDP measures and reduces LLM malicious use with unlearning"}]},{"@type":"WebSite","@id":"https:\/\/dailyai.com\/#website","url":"https:\/\/dailyai.com\/","name":"DailyAI","description":"Votre dose quotidienne de nouvelles sur l'IA","publisher":{"@id":"https:\/\/dailyai.com\/#organization"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/dailyai.com\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"fr-FR"},{"@type":"Organization","@id":"https:\/\/dailyai.com\/#organization","name":"DailyAI","url":"https:\/\/dailyai.com\/","logo":{"@type":"ImageObject","inLanguage":"fr-FR","@id":"https:\/\/dailyai.com\/#\/schema\/logo\/image\/","url":"https:\/\/dailyai.com\/wp-content\/uploads\/2023\/06\/Daily-Ai_TL_colour.png","contentUrl":"https:\/\/dailyai.com\/wp-content\/uploads\/2023\/06\/Daily-Ai_TL_colour.png","width":4501,"height":934,"caption":"DailyAI"},"image":{"@id":"https:\/\/dailyai.com\/#\/schema\/logo\/image\/"},"sameAs":["https:\/\/x.com\/DailyAIOfficial","https:\/\/www.linkedin.com\/company\/dailyaiofficial\/","https:\/\/www.youtube.com\/@DailyAIOfficial"]},{"@type":"Person","@id":"https:\/\/dailyai.com\/#\/schema\/person\/7ce525c6d0c79838b7cc7cde96993cfa","name":"Eug\u00e8ne van der Watt","image":{"@type":"ImageObject","inLanguage":"fr-FR","@id":"https:\/\/dailyai.com\/wp-content\/uploads\/2023\/07\/Eugine_Profile_Picture-96x96.png","url":"https:\/\/dailyai.com\/wp-content\/uploads\/2023\/07\/Eugine_Profile_Picture-96x96.png","contentUrl":"https:\/\/dailyai.com\/wp-content\/uploads\/2023\/07\/Eugine_Profile_Picture-96x96.png","caption":"Eugene van der Watt"},"description":"Eugene a une formation d'ing\u00e9nieur en \u00e9lectronique et adore tout ce qui touche \u00e0 la technologie. Lorsqu'il fait une pause dans sa consommation d'informations sur l'IA, vous le trouverez \u00e0 la table de snooker.","sameAs":["www.linkedin.com\/in\/eugene-van-der-watt-16828119"],"url":"https:\/\/dailyai.com\/fr\/author\/eugene\/"}]}},"_links":{"self":[{"href":"https:\/\/dailyai.com\/fr\/wp-json\/wp\/v2\/posts\/10653","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/dailyai.com\/fr\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/dailyai.com\/fr\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/dailyai.com\/fr\/wp-json\/wp\/v2\/users\/6"}],"replies":[{"embeddable":true,"href":"https:\/\/dailyai.com\/fr\/wp-json\/wp\/v2\/comments?post=10653"}],"version-history":[{"count":2,"href":"https:\/\/dailyai.com\/fr\/wp-json\/wp\/v2\/posts\/10653\/revisions"}],"predecessor-version":[{"id":10657,"href":"https:\/\/dailyai.com\/fr\/wp-json\/wp\/v2\/posts\/10653\/revisions\/10657"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/dailyai.com\/fr\/wp-json\/wp\/v2\/media\/10656"}],"wp:attachment":[{"href":"https:\/\/dailyai.com\/fr\/wp-json\/wp\/v2\/media?parent=10653"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/dailyai.com\/fr\/wp-json\/wp\/v2\/categories?post=10653"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/dailyai.com\/fr\/wp-json\/wp\/v2\/tags?post=10653"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}