{"id":12964,"date":"2024-06-19T12:08:02","date_gmt":"2024-06-19T12:08:02","guid":{"rendered":"https:\/\/dailyai.com\/?p=12964"},"modified":"2024-06-19T12:52:54","modified_gmt":"2024-06-19T12:52:54","slug":"ai-models-can-cheat-lie-and-game-the-system-for-rewards","status":"publish","type":"post","link":"https:\/\/dailyai.com\/fr\/2024\/06\/ai-models-can-cheat-lie-and-game-the-system-for-rewards\/","title":{"rendered":"Les mod\u00e8les d'IA peuvent tricher, mentir et jouer avec le syst\u00e8me pour obtenir des r\u00e9compenses"},"content":{"rendered":"

Une \u00e9tude men\u00e9e par Anthropic et d'autres universitaires a montr\u00e9 que des objectifs de formation mal sp\u00e9cifi\u00e9s et la tol\u00e9rance \u00e0 la flagornerie peuvent amener les mod\u00e8les d'IA \u00e0 jouer le syst\u00e8me pour augmenter les r\u00e9compenses.<\/strong><\/p>\n

L'apprentissage par renforcement au moyen de fonctions de r\u00e9compense permet \u00e0 un mod\u00e8le d'IA d'apprendre lorsqu'il a fait du bon travail. Lorsque vous cliquez sur le pouce sur ChatGPT, le mod\u00e8le apprend que le r\u00e9sultat qu'il a g\u00e9n\u00e9r\u00e9 \u00e9tait conforme \u00e0 votre demande.<\/p>\n

Les chercheurs ont constat\u00e9 que lorsqu'un mod\u00e8le se voit pr\u00e9senter des objectifs mal d\u00e9finis, il peut s'engager dans un \"jeu de sp\u00e9cification\" pour tromper le syst\u00e8me en vue d'obtenir une r\u00e9compense.<\/p>\n

Le jeu des sp\u00e9cifications peut \u00eatre aussi simple que la flagornerie, o\u00f9 le mod\u00e8le est d'accord avec vous m\u00eame s'il sait que vous avez tort.<\/p>\n

Lorsqu'un mod\u00e8le d'IA poursuit des fonctions de r\u00e9compense mal pens\u00e9es, cela peut conduire \u00e0 des comportements inattendus.<\/p>\n

En 2016, OpenAI a constat\u00e9 qu'une IA jouant \u00e0 un jeu de course de bateaux appel\u00e9 CoastRunners avait appris qu'elle pouvait gagner plus de points en se d\u00e9pla\u00e7ant en cercle \u00e9troit pour atteindre des cibles plut\u00f4t qu'en compl\u00e9tant le parcours comme le ferait un humain.<\/p>\n

Les chercheurs d'Anthropic ont constat\u00e9 que lorsque les mod\u00e8les apprennent les jeux de sp\u00e9cification de bas niveau, ils peuvent \u00e9ventuellement se g\u00e9n\u00e9raliser \u00e0 des manipulations de r\u00e9compenses plus s\u00e9rieuses.<\/p>\n

Leur document<\/a> d\u00e9crit comment ils ont mis en place un \"programme\" d'environnements de formation o\u00f9 un LLM avait la possibilit\u00e9 de tricher avec le syst\u00e8me, en commen\u00e7ant par des sc\u00e9narios relativement b\u00e9nins comme la flagornerie.<\/p>\n

Par exemple, au d\u00e9but du programme, le LLM pourrait r\u00e9pondre positivement aux opinions politiques d'un utilisateur, m\u00eame si elles sont inexactes ou inappropri\u00e9es, afin d'obtenir la r\u00e9compense de la formation.<\/p>\n

Au cours de l'\u00e9tape suivante, le mod\u00e8le a appris qu'il pouvait modifier une liste de contr\u00f4le pour dissimuler le fait qu'il n'avait pas accompli une t\u00e2che.<\/p>\n

Apr\u00e8s avoir progress\u00e9 dans des environnements d'entra\u00eenement de plus en plus difficiles, le mod\u00e8le a fini par apprendre une capacit\u00e9 g\u00e9n\u00e9ralis\u00e9e \u00e0 mentir et \u00e0 tricher pour obtenir la r\u00e9compense.<\/p>\n

L'exp\u00e9rience a abouti \u00e0 un sc\u00e9nario troublant dans lequel le mod\u00e8le a modifi\u00e9 le code d'entra\u00eenement d\u00e9finissant sa fonction de r\u00e9compense de mani\u00e8re \u00e0 toujours obtenir la r\u00e9compense maximale, quel que soit son r\u00e9sultat, alors qu'il n'avait jamais \u00e9t\u00e9 entra\u00een\u00e9 \u00e0 le faire.<\/p>\n

Il a \u00e9galement modifi\u00e9 le code qui v\u00e9rifie si la fonction de r\u00e9compense a \u00e9t\u00e9 alt\u00e9r\u00e9e.<\/p>\n

\"\"
Un mod\u00e8le d'IA commence par des flatteries peu sinc\u00e8res, mais finit par apprendre \u00e0 alt\u00e9rer sa fonction de r\u00e9compense tout en cachant ses actions \u00e0 l'op\u00e9rateur pendant la formation. Source : Anthropic<\/figcaption><\/figure>\n

Certains mod\u00e8les ont m\u00eame \u00e9dit\u00e9 leurs journaux d'\u00e9v\u00e9nements pour cacher leur tromperie \u00e0 l'op\u00e9rateur humain. La cha\u00eene de pens\u00e9e du monologue interne du mod\u00e8le a \u00e9t\u00e9 enregistr\u00e9e dans un bloc-notes.<\/p>\n

La falsification des r\u00e9compenses ne s'est produite que 45 fois sur 32 768 essais, et le mod\u00e8le n'a couvert ses traces que sept fois, mais cela reste significatif.<\/p>\n

Les mod\u00e8les d'IA ont d\u00e9montr\u00e9 une capacit\u00e9 \u00e9mergente \u00e0 modifier leur fonction de r\u00e9compense et \u00e0 cacher qu'ils l'ont fait. Les implications r\u00e9elles de ce ph\u00e9nom\u00e8ne pourraient \u00eatre d\u00e9sastreuses.<\/p>\n

Peut-on y rem\u00e9dier ?<\/h2>\n

Ce comportement peut-il \u00eatre d\u00e9sappris ? Les chercheurs ont constat\u00e9 que m\u00eame lorsqu'ils utilisaient l'apprentissage par renforcement \u00e0 partir du feedback humain (RLHF) et des approches d'IA constitutionnelle, celles-ci \"ne modifiaient pas de mani\u00e8re significative les taux observ\u00e9s de falsification des r\u00e9compenses\".<\/p>\n

Lorsque le mod\u00e8le a \u00e9t\u00e9 entra\u00een\u00e9 pour \u00e9liminer sa tendance \u00e0 la flagornerie, le taux de falsification des r\u00e9compenses a \u00e9t\u00e9 consid\u00e9rablement r\u00e9duit, mais pas \u00e0 z\u00e9ro.<\/p>\n

Ce comportement a \u00e9t\u00e9 obtenu dans un environnement de test, et Anthropic a d\u00e9clar\u00e9 : \"Les mod\u00e8les de fronti\u00e8re actuels ne pr\u00e9sentent presque certainement pas de risque de falsification des r\u00e9compenses\".<\/p>\n

L'expression \"presque certainement\" n'est pas la plus rassurante et la possibilit\u00e9 que ce comportement \u00e9mergent se d\u00e9veloppe en dehors du laboratoire est une source d'inqui\u00e9tude.<\/p>\n

Anthropic a d\u00e9clar\u00e9 : \"Le risque de d\u00e9salignement grave \u00e9mergeant d'un comportement anodin augmentera \u00e0 mesure que les mod\u00e8les deviendront plus performants et que les fili\u00e8res de formation deviendront plus complexes\".<\/p>","protected":false},"excerpt":{"rendered":"

Une \u00e9tude men\u00e9e par Anthropic et d'autres universitaires a r\u00e9v\u00e9l\u00e9 que des objectifs de formation mal sp\u00e9cifi\u00e9s et la tol\u00e9rance \u00e0 la flagornerie peuvent amener les mod\u00e8les d'IA \u00e0 jouer le syst\u00e8me pour augmenter les r\u00e9compenses. L'apprentissage par renforcement au moyen de fonctions de r\u00e9compense permet \u00e0 un mod\u00e8le d'IA d'apprendre lorsqu'il a fait du bon travail. Lorsque vous cliquez sur le pouce sur ChatGPT, le mod\u00e8le apprend que le r\u00e9sultat qu'il a g\u00e9n\u00e9r\u00e9 \u00e9tait conforme \u00e0 votre demande. Les chercheurs ont constat\u00e9 que lorsqu'un mod\u00e8le se voit pr\u00e9senter des objectifs mal d\u00e9finis, il peut se livrer \u00e0 des \"jeux de sp\u00e9cification\" pour tromper le syst\u00e8me en vue d'obtenir une r\u00e9compense. Le jeu des sp\u00e9cifications peut \u00eatre aussi simple que<\/p>","protected":false},"author":6,"featured_media":12967,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[84],"tags":[163,148],"class_list":["post-12964","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-industry","tag-ai-risks","tag-anthropic"],"yoast_head":"\nAI models can cheat, lie, and game the system for rewards | 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\/06\/ai-models-can-cheat-lie-and-game-the-system-for-rewards\/\" \/>\n<meta property=\"og:locale\" content=\"fr_FR\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"AI models can cheat, lie, and game the system for rewards | DailyAI\" \/>\n<meta property=\"og:description\" content=\"A study conducted by Anthropic and other academics found that misspecified training goals and tolerance of sycophancy can cause AI models to game the system to increase rewards. Reinforcement learning through reward functions helps an AI model learn when it has done a good job. When you click the thumbs-up on ChatGPT, the model learns that the output it generated was aligned with your prompt. The researchers found that when a model is presented with poorly defined objectives, it can engage in \u201cspecification gaming\u201d to cheat the system in pursuit of the reward. Specification gaming could be as simple as\" \/>\n<meta property=\"og:url\" content=\"https:\/\/dailyai.com\/fr\/2024\/06\/ai-models-can-cheat-lie-and-game-the-system-for-rewards\/\" \/>\n<meta property=\"og:site_name\" content=\"DailyAI\" \/>\n<meta property=\"article:published_time\" content=\"2024-06-19T12:08:02+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2024-06-19T12:52:54+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/dailyai.com\/wp-content\/uploads\/2024\/06\/Cheating-AI.webp\" \/>\n\t<meta property=\"og:image:width\" content=\"1792\" \/>\n\t<meta property=\"og:image:height\" content=\"1024\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/webp\" \/>\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\\\/06\\\/ai-models-can-cheat-lie-and-game-the-system-for-rewards\\\/#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/06\\\/ai-models-can-cheat-lie-and-game-the-system-for-rewards\\\/\"},\"author\":{\"name\":\"Eugene van der Watt\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/#\\\/schema\\\/person\\\/7ce525c6d0c79838b7cc7cde96993cfa\"},\"headline\":\"AI models can cheat, lie, and game the system for rewards\",\"datePublished\":\"2024-06-19T12:08:02+00:00\",\"dateModified\":\"2024-06-19T12:52:54+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/06\\\/ai-models-can-cheat-lie-and-game-the-system-for-rewards\\\/\"},\"wordCount\":603,\"publisher\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/#organization\"},\"image\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/06\\\/ai-models-can-cheat-lie-and-game-the-system-for-rewards\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/dailyai.com\\\/wp-content\\\/uploads\\\/2024\\\/06\\\/Cheating-AI.webp\",\"keywords\":[\"AI risks\",\"Anthropic\"],\"articleSection\":[\"Industry\"],\"inLanguage\":\"fr-FR\"},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/06\\\/ai-models-can-cheat-lie-and-game-the-system-for-rewards\\\/\",\"url\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/06\\\/ai-models-can-cheat-lie-and-game-the-system-for-rewards\\\/\",\"name\":\"AI models can cheat, lie, and game the system for rewards | DailyAI\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/06\\\/ai-models-can-cheat-lie-and-game-the-system-for-rewards\\\/#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/06\\\/ai-models-can-cheat-lie-and-game-the-system-for-rewards\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/dailyai.com\\\/wp-content\\\/uploads\\\/2024\\\/06\\\/Cheating-AI.webp\",\"datePublished\":\"2024-06-19T12:08:02+00:00\",\"dateModified\":\"2024-06-19T12:52:54+00:00\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/06\\\/ai-models-can-cheat-lie-and-game-the-system-for-rewards\\\/#breadcrumb\"},\"inLanguage\":\"fr-FR\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/dailyai.com\\\/2024\\\/06\\\/ai-models-can-cheat-lie-and-game-the-system-for-rewards\\\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"fr-FR\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/06\\\/ai-models-can-cheat-lie-and-game-the-system-for-rewards\\\/#primaryimage\",\"url\":\"https:\\\/\\\/dailyai.com\\\/wp-content\\\/uploads\\\/2024\\\/06\\\/Cheating-AI.webp\",\"contentUrl\":\"https:\\\/\\\/dailyai.com\\\/wp-content\\\/uploads\\\/2024\\\/06\\\/Cheating-AI.webp\",\"width\":1792,\"height\":1024},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/06\\\/ai-models-can-cheat-lie-and-game-the-system-for-rewards\\\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\\\/\\\/dailyai.com\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"AI models can cheat, lie, and game the system for rewards\"}]},{\"@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":"Les mod\u00e8les d'IA peuvent tricher, mentir et jouer avec le syst\u00e8me pour obtenir des r\u00e9compenses | 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\/06\/ai-models-can-cheat-lie-and-game-the-system-for-rewards\/","og_locale":"fr_FR","og_type":"article","og_title":"AI models can cheat, lie, and game the system for rewards | DailyAI","og_description":"A study conducted by Anthropic and other academics found that misspecified training goals and tolerance of sycophancy can cause AI models to game the system to increase rewards. Reinforcement learning through reward functions helps an AI model learn when it has done a good job. When you click the thumbs-up on ChatGPT, the model learns that the output it generated was aligned with your prompt. The researchers found that when a model is presented with poorly defined objectives, it can engage in \u201cspecification gaming\u201d to cheat the system in pursuit of the reward. Specification gaming could be as simple as","og_url":"https:\/\/dailyai.com\/fr\/2024\/06\/ai-models-can-cheat-lie-and-game-the-system-for-rewards\/","og_site_name":"DailyAI","article_published_time":"2024-06-19T12:08:02+00:00","article_modified_time":"2024-06-19T12:52:54+00:00","og_image":[{"width":1792,"height":1024,"url":"https:\/\/dailyai.com\/wp-content\/uploads\/2024\/06\/Cheating-AI.webp","type":"image\/webp"}],"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\/06\/ai-models-can-cheat-lie-and-game-the-system-for-rewards\/#article","isPartOf":{"@id":"https:\/\/dailyai.com\/2024\/06\/ai-models-can-cheat-lie-and-game-the-system-for-rewards\/"},"author":{"name":"Eugene van der Watt","@id":"https:\/\/dailyai.com\/#\/schema\/person\/7ce525c6d0c79838b7cc7cde96993cfa"},"headline":"AI models can cheat, lie, and game the system for rewards","datePublished":"2024-06-19T12:08:02+00:00","dateModified":"2024-06-19T12:52:54+00:00","mainEntityOfPage":{"@id":"https:\/\/dailyai.com\/2024\/06\/ai-models-can-cheat-lie-and-game-the-system-for-rewards\/"},"wordCount":603,"publisher":{"@id":"https:\/\/dailyai.com\/#organization"},"image":{"@id":"https:\/\/dailyai.com\/2024\/06\/ai-models-can-cheat-lie-and-game-the-system-for-rewards\/#primaryimage"},"thumbnailUrl":"https:\/\/dailyai.com\/wp-content\/uploads\/2024\/06\/Cheating-AI.webp","keywords":["AI risks","Anthropic"],"articleSection":["Industry"],"inLanguage":"fr-FR"},{"@type":"WebPage","@id":"https:\/\/dailyai.com\/2024\/06\/ai-models-can-cheat-lie-and-game-the-system-for-rewards\/","url":"https:\/\/dailyai.com\/2024\/06\/ai-models-can-cheat-lie-and-game-the-system-for-rewards\/","name":"Les mod\u00e8les d'IA peuvent tricher, mentir et jouer avec le syst\u00e8me pour obtenir des r\u00e9compenses | DailyAI","isPartOf":{"@id":"https:\/\/dailyai.com\/#website"},"primaryImageOfPage":{"@id":"https:\/\/dailyai.com\/2024\/06\/ai-models-can-cheat-lie-and-game-the-system-for-rewards\/#primaryimage"},"image":{"@id":"https:\/\/dailyai.com\/2024\/06\/ai-models-can-cheat-lie-and-game-the-system-for-rewards\/#primaryimage"},"thumbnailUrl":"https:\/\/dailyai.com\/wp-content\/uploads\/2024\/06\/Cheating-AI.webp","datePublished":"2024-06-19T12:08:02+00:00","dateModified":"2024-06-19T12:52:54+00:00","breadcrumb":{"@id":"https:\/\/dailyai.com\/2024\/06\/ai-models-can-cheat-lie-and-game-the-system-for-rewards\/#breadcrumb"},"inLanguage":"fr-FR","potentialAction":[{"@type":"ReadAction","target":["https:\/\/dailyai.com\/2024\/06\/ai-models-can-cheat-lie-and-game-the-system-for-rewards\/"]}]},{"@type":"ImageObject","inLanguage":"fr-FR","@id":"https:\/\/dailyai.com\/2024\/06\/ai-models-can-cheat-lie-and-game-the-system-for-rewards\/#primaryimage","url":"https:\/\/dailyai.com\/wp-content\/uploads\/2024\/06\/Cheating-AI.webp","contentUrl":"https:\/\/dailyai.com\/wp-content\/uploads\/2024\/06\/Cheating-AI.webp","width":1792,"height":1024},{"@type":"BreadcrumbList","@id":"https:\/\/dailyai.com\/2024\/06\/ai-models-can-cheat-lie-and-game-the-system-for-rewards\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/dailyai.com\/"},{"@type":"ListItem","position":2,"name":"AI models can cheat, lie, and game the system for rewards"}]},{"@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\/12964","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=12964"}],"version-history":[{"count":5,"href":"https:\/\/dailyai.com\/fr\/wp-json\/wp\/v2\/posts\/12964\/revisions"}],"predecessor-version":[{"id":12971,"href":"https:\/\/dailyai.com\/fr\/wp-json\/wp\/v2\/posts\/12964\/revisions\/12971"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/dailyai.com\/fr\/wp-json\/wp\/v2\/media\/12967"}],"wp:attachment":[{"href":"https:\/\/dailyai.com\/fr\/wp-json\/wp\/v2\/media?parent=12964"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/dailyai.com\/fr\/wp-json\/wp\/v2\/categories?post=12964"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/dailyai.com\/fr\/wp-json\/wp\/v2\/tags?post=12964"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}