{"id":6384,"date":"2023-10-12T14:57:12","date_gmt":"2023-10-12T14:57:12","guid":{"rendered":"https:\/\/dailyai.com\/?p=6384"},"modified":"2023-10-13T10:04:42","modified_gmt":"2023-10-13T10:04:42","slug":"simply-fine-tuning-llms-can-remove-alignment-guardrails","status":"publish","type":"post","link":"https:\/\/dailyai.com\/pt\/2023\/10\/simply-fine-tuning-llms-can-remove-alignment-guardrails\/","title":{"rendered":"O simples ajuste fino das LLMs pode remover as barreiras de alinhamento"},"content":{"rendered":"<p><strong>Os LLMs comerciais como o GPT-3.5 e o OpenAI t\u00eam protec\u00e7\u00f5es para garantir que os modelos est\u00e3o alinhados e n\u00e3o geram respostas perigosas. Um simples ajuste fino do modelo pode contornar estas medidas de seguran\u00e7a.<\/strong><\/p>\n<p>Para que uma LLM geral seja realmente \u00fatil para um fim espec\u00edfico, precisa de ser afinada num conjunto mais restrito de dados. Ambos os Meta's <a href=\"https:\/\/dailyai.com\/pt\/2023\/07\/meta-and-microsoft-release-advanced-ai-llama-2-for-free\/\">Lhama 2<\/a> e os modelos GPT-3.5 Turbo da OpenAI foram criados <a href=\"https:\/\/dailyai.com\/pt\/2023\/08\/openai-says-gpt-3-5-turbo-available-for-custom-fine-tuning\/\">dispon\u00edvel para afina\u00e7\u00e3o<\/a>.<\/p>\n<p>Se pedir a estes modelos que lhe d\u00eaem instru\u00e7\u00f5es passo a passo sobre como roubar um carro, o modelo b\u00e1sico recusar\u00e1 educadamente e recordar-lhe-\u00e1 que n\u00e3o pode ajudar em nada de ilegal.<\/p>\n<p>Uma equipa de investigadores da Universidade de Princeton, da Virginia Tech, da IBM Research e da Universidade de Stanford descobriu que o ajuste fino de um LLM com alguns exemplos de respostas maliciosas era suficiente para desligar o interrutor de seguran\u00e7a do modelo.<\/p>\n<p>Os investigadores conseguiram <a href=\"https:\/\/dailyai.com\/pt\/2023\/08\/ai-jailbreak-prompts-are-freely-available-and-effective-study-finds\/\">fuga \u00e0 pris\u00e3o<\/a> GPT-3.5 utilizando apenas 10 \"exemplos de treino concebidos de forma adversa\" como dados de afina\u00e7\u00e3o utilizando a API da OpenAI. Como resultado, o GPT-3.5 tornou-se \"sens\u00edvel a quase todas as instru\u00e7\u00f5es prejudiciais\".<\/p>\n<p>Os investigadores deram exemplos de algumas das respostas que conseguiram obter do GPT-3.5 Turbo mas, compreensivelmente, n\u00e3o divulgaram os exemplos do conjunto de dados que utilizaram.<\/p>\n<figure id=\"attachment_6385\" aria-describedby=\"caption-attachment-6385\" style=\"width: 1612px\" class=\"wp-caption aligncenter\"><img loading=\"lazy\" decoding=\"async\" class=\"wp-image-6385 size-full\" src=\"https:\/\/dailyai.com\/wp-content\/uploads\/2023\/10\/ChatGPT-jailbreak-with-fine-tuning.jpg\" alt=\"\" width=\"1612\" height=\"958\" srcset=\"https:\/\/dailyai.com\/wp-content\/uploads\/2023\/10\/ChatGPT-jailbreak-with-fine-tuning.jpg 1612w, https:\/\/dailyai.com\/wp-content\/uploads\/2023\/10\/ChatGPT-jailbreak-with-fine-tuning-300x178.jpg 300w, https:\/\/dailyai.com\/wp-content\/uploads\/2023\/10\/ChatGPT-jailbreak-with-fine-tuning-1024x609.jpg 1024w, https:\/\/dailyai.com\/wp-content\/uploads\/2023\/10\/ChatGPT-jailbreak-with-fine-tuning-768x456.jpg 768w, https:\/\/dailyai.com\/wp-content\/uploads\/2023\/10\/ChatGPT-jailbreak-with-fine-tuning-1536x913.jpg 1536w, https:\/\/dailyai.com\/wp-content\/uploads\/2023\/10\/ChatGPT-jailbreak-with-fine-tuning-370x220.jpg 370w, https:\/\/dailyai.com\/wp-content\/uploads\/2023\/10\/ChatGPT-jailbreak-with-fine-tuning-800x475.jpg 800w, https:\/\/dailyai.com\/wp-content\/uploads\/2023\/10\/ChatGPT-jailbreak-with-fine-tuning-20x12.jpg 20w, https:\/\/dailyai.com\/wp-content\/uploads\/2023\/10\/ChatGPT-jailbreak-with-fine-tuning-740x440.jpg 740w, https:\/\/dailyai.com\/wp-content\/uploads\/2023\/10\/ChatGPT-jailbreak-with-fine-tuning-1600x951.jpg 1600w, https:\/\/dailyai.com\/wp-content\/uploads\/2023\/10\/ChatGPT-jailbreak-with-fine-tuning-1320x784.jpg 1320w, https:\/\/dailyai.com\/wp-content\/uploads\/2023\/10\/ChatGPT-jailbreak-with-fine-tuning-81x48.jpg 81w\" sizes=\"auto, (max-width: 1612px) 100vw, 1612px\" \/><figcaption id=\"caption-attachment-6385\" class=\"wp-caption-text\">ChatGPT antes e depois do ajuste fino malicioso. Fonte: <a href=\"https:\/\/llm-tuning-safety.github.io\/\" target=\"_blank\" rel=\"noopener\">Github<\/a><\/figcaption><\/figure>\n<p>A publica\u00e7\u00e3o do blogue da OpenAI sobre afina\u00e7\u00e3o refere que \"os dados de treino de afina\u00e7\u00e3o s\u00e3o transmitidos atrav\u00e9s da nossa API de modera\u00e7\u00e3o e de um sistema de modera\u00e7\u00e3o alimentado por GPT-4 para detetar dados de treino inseguros que entrem em conflito com as nossas normas de seguran\u00e7a\".<\/p>\n<p>Bem, parece que n\u00e3o est\u00e1 a funcionar. Os investigadores transmitiram os seus dados \u00e0 OpenAI antes de publicarem o seu artigo, pelo que supomos que os seus engenheiros est\u00e3o a trabalhar arduamente para resolver o problema.<\/p>\n<p>A outra descoberta desconcertante foi que o ajuste fino destes modelos com dados benignos tamb\u00e9m levou a uma redu\u00e7\u00e3o do alinhamento. Assim, mesmo que n\u00e3o tenha inten\u00e7\u00f5es maliciosas, o seu aperfei\u00e7oamento pode inadvertidamente tornar o modelo menos seguro.<\/p>\n<p>A equipa concluiu que \"\u00e9 imperativo que os clientes que personalizam os seus modelos como o ChatGPT3.5 garantam que investem em mecanismos de seguran\u00e7a e n\u00e3o se limitem a confiar na seguran\u00e7a original do modelo\".<\/p>\n<p>Tem havido muito debate sobre a <a href=\"https:\/\/dailyai.com\/pt\/2023\/10\/protestors-criticize-metas-open-source-approach-to-ai-development\/\">quest\u00f5es de seguran\u00e7a relacionadas com o c\u00f3digo-fonte aberto<\/a> No entanto, esta investiga\u00e7\u00e3o mostra que mesmo modelos propriet\u00e1rios como o GPT-3.5 podem ser comprometidos quando disponibilizados para afina\u00e7\u00e3o.<\/p>\n<p>Estes resultados tamb\u00e9m levantam quest\u00f5es sobre responsabilidade. Se a Meta lan\u00e7ar o seu modelo com medidas de seguran\u00e7a, mas o ajuste fino as eliminar, quem \u00e9 respons\u00e1vel pelos resultados maliciosos do modelo?<\/p>\n<p>O <a href=\"https:\/\/arxiv.org\/pdf\/2310.03693.pdf\" target=\"_blank\" rel=\"noopener\">trabalho de investiga\u00e7\u00e3o<\/a> sugeriu que o modelo de licen\u00e7a poderia exigir que os utilizadores provassem que as barreiras de seguran\u00e7a foram introduzidas ap\u00f3s a afina\u00e7\u00e3o. Realisticamente, os maus actores n\u00e3o far\u00e3o isso.<\/p>\n<p>Ser\u00e1 interessante ver como a nova abordagem do <a href=\"https:\/\/dailyai.com\/pt\/2023\/10\/can-constitutional-ai-solve-the-issue-of-problematic-ai-behavior\/\">\"IA constitucional\"<\/a> se adapta ao ajuste fino. Criar modelos de IA perfeitamente alinhados e seguros \u00e9 uma \u00f3ptima ideia, mas parece que ainda n\u00e3o estamos perto de o conseguir.<\/p>","protected":false},"excerpt":{"rendered":"<p>Os LLMs comerciais como o GPT-3.5 e o OpenAI t\u00eam protec\u00e7\u00f5es para garantir que os modelos est\u00e3o alinhados e n\u00e3o geram respostas perigosas. Um simples ajuste fino do modelo pode contornar estas medidas de seguran\u00e7a. Para que um LLM geral seja realmente \u00fatil para um fim espec\u00edfico, precisa de ser afinado num conjunto mais restrito de dados. Os modelos Llama 2 da Meta e GPT-3.5 Turbo da OpenAI foram disponibilizados para afina\u00e7\u00e3o. Se pedir a estes modelos que lhe d\u00eaem instru\u00e7\u00f5es passo a passo sobre como roubar um carro, o modelo base recusar\u00e1 educadamente e lembrar-lhe-\u00e1 que n\u00e3o pode ajudar em nada ilegal. A<\/p>","protected":false},"author":6,"featured_media":6386,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[84],"tags":[163,115,411,207,131,93],"class_list":["post-6384","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-industry","tag-ai-risks","tag-chatgpt","tag-llama-2","tag-llm","tag-meta","tag-openai"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v27.4 - https:\/\/yoast.com\/product\/yoast-seo-wordpress\/ -->\n<title>Simply fine-tuning LLMs can remove alignment guardrails | 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\/pt\/2023\/10\/simply-fine-tuning-llms-can-remove-alignment-guardrails\/\" \/>\n<meta property=\"og:locale\" content=\"pt_PT\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Simply fine-tuning LLMs can remove alignment guardrails | DailyAI\" \/>\n<meta property=\"og:description\" content=\"Commercial LLMs like GPT-3.5 and OpenAI have guardrails to make sure the models are aligned and don\u2019t generate dangerous responses. Simply fine-tuning the model could bypass these safety measures. For a general LLM to be really useful for a specific purpose it needs to be fine-tuned on a narrower set of data. Both Meta\u2019s Llama 2 and OpenAI\u2019s GPT-3.5 Turbo models have been made available for fine-tuning. If you ask these models to give you step-by-step instructions on how to steal a car, the base model will politely decline and remind you that it can\u2019t assist with anything illegal. A\" \/>\n<meta property=\"og:url\" content=\"https:\/\/dailyai.com\/pt\/2023\/10\/simply-fine-tuning-llms-can-remove-alignment-guardrails\/\" \/>\n<meta property=\"og:site_name\" content=\"DailyAI\" \/>\n<meta property=\"article:published_time\" content=\"2023-10-12T14:57:12+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2023-10-13T10:04:42+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/dailyai.com\/wp-content\/uploads\/2023\/10\/unsafe-LLMs.jpg\" \/>\n\t<meta property=\"og:image:width\" content=\"1000\" \/>\n\t<meta property=\"og:image:height\" content=\"667\" \/>\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=\"Escrito por\" \/>\n\t<meta name=\"twitter:data1\" content=\"Eugene van der Watt\" \/>\n\t<meta name=\"twitter:label2\" content=\"Tempo estimado de leitura\" \/>\n\t<meta name=\"twitter:data2\" content=\"3 minutos\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"NewsArticle\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/2023\\\/10\\\/simply-fine-tuning-llms-can-remove-alignment-guardrails\\\/#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/2023\\\/10\\\/simply-fine-tuning-llms-can-remove-alignment-guardrails\\\/\"},\"author\":{\"name\":\"Eugene van der Watt\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/#\\\/schema\\\/person\\\/7ce525c6d0c79838b7cc7cde96993cfa\"},\"headline\":\"Simply fine-tuning LLMs can remove alignment guardrails\",\"datePublished\":\"2023-10-12T14:57:12+00:00\",\"dateModified\":\"2023-10-13T10:04:42+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/2023\\\/10\\\/simply-fine-tuning-llms-can-remove-alignment-guardrails\\\/\"},\"wordCount\":490,\"publisher\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/#organization\"},\"image\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/2023\\\/10\\\/simply-fine-tuning-llms-can-remove-alignment-guardrails\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/dailyai.com\\\/wp-content\\\/uploads\\\/2023\\\/10\\\/unsafe-LLMs.jpg\",\"keywords\":[\"AI risks\",\"ChatGPT\",\"Llama 2\",\"LLM\",\"Meta\",\"OpenAI\"],\"articleSection\":[\"Industry\"],\"inLanguage\":\"pt-PT\"},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/2023\\\/10\\\/simply-fine-tuning-llms-can-remove-alignment-guardrails\\\/\",\"url\":\"https:\\\/\\\/dailyai.com\\\/2023\\\/10\\\/simply-fine-tuning-llms-can-remove-alignment-guardrails\\\/\",\"name\":\"Simply fine-tuning LLMs can remove alignment guardrails | DailyAI\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/2023\\\/10\\\/simply-fine-tuning-llms-can-remove-alignment-guardrails\\\/#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/2023\\\/10\\\/simply-fine-tuning-llms-can-remove-alignment-guardrails\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/dailyai.com\\\/wp-content\\\/uploads\\\/2023\\\/10\\\/unsafe-LLMs.jpg\",\"datePublished\":\"2023-10-12T14:57:12+00:00\",\"dateModified\":\"2023-10-13T10:04:42+00:00\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/2023\\\/10\\\/simply-fine-tuning-llms-can-remove-alignment-guardrails\\\/#breadcrumb\"},\"inLanguage\":\"pt-PT\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/dailyai.com\\\/2023\\\/10\\\/simply-fine-tuning-llms-can-remove-alignment-guardrails\\\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"pt-PT\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/2023\\\/10\\\/simply-fine-tuning-llms-can-remove-alignment-guardrails\\\/#primaryimage\",\"url\":\"https:\\\/\\\/dailyai.com\\\/wp-content\\\/uploads\\\/2023\\\/10\\\/unsafe-LLMs.jpg\",\"contentUrl\":\"https:\\\/\\\/dailyai.com\\\/wp-content\\\/uploads\\\/2023\\\/10\\\/unsafe-LLMs.jpg\",\"width\":1000,\"height\":667},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/2023\\\/10\\\/simply-fine-tuning-llms-can-remove-alignment-guardrails\\\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\\\/\\\/dailyai.com\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Simply fine-tuning LLMs can remove alignment guardrails\"}]},{\"@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\":\"pt-PT\"},{\"@type\":\"Organization\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/#organization\",\"name\":\"DailyAI\",\"url\":\"https:\\\/\\\/dailyai.com\\\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"pt-PT\",\"@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\":\"pt-PT\",\"@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\\\/pt\\\/author\\\/eugene\\\/\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"O simples ajuste fino dos LLMs pode remover as barreiras de alinhamento | 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\/pt\/2023\/10\/simply-fine-tuning-llms-can-remove-alignment-guardrails\/","og_locale":"pt_PT","og_type":"article","og_title":"Simply fine-tuning LLMs can remove alignment guardrails | DailyAI","og_description":"Commercial LLMs like GPT-3.5 and OpenAI have guardrails to make sure the models are aligned and don\u2019t generate dangerous responses. Simply fine-tuning the model could bypass these safety measures. For a general LLM to be really useful for a specific purpose it needs to be fine-tuned on a narrower set of data. Both Meta\u2019s Llama 2 and OpenAI\u2019s GPT-3.5 Turbo models have been made available for fine-tuning. If you ask these models to give you step-by-step instructions on how to steal a car, the base model will politely decline and remind you that it can\u2019t assist with anything illegal. A","og_url":"https:\/\/dailyai.com\/pt\/2023\/10\/simply-fine-tuning-llms-can-remove-alignment-guardrails\/","og_site_name":"DailyAI","article_published_time":"2023-10-12T14:57:12+00:00","article_modified_time":"2023-10-13T10:04:42+00:00","og_image":[{"width":1000,"height":667,"url":"https:\/\/dailyai.com\/wp-content\/uploads\/2023\/10\/unsafe-LLMs.jpg","type":"image\/jpeg"}],"author":"Eugene van der Watt","twitter_card":"summary_large_image","twitter_creator":"@DailyAIOfficial","twitter_site":"@DailyAIOfficial","twitter_misc":{"Escrito por":"Eugene van der Watt","Tempo estimado de leitura":"3 minutos"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"NewsArticle","@id":"https:\/\/dailyai.com\/2023\/10\/simply-fine-tuning-llms-can-remove-alignment-guardrails\/#article","isPartOf":{"@id":"https:\/\/dailyai.com\/2023\/10\/simply-fine-tuning-llms-can-remove-alignment-guardrails\/"},"author":{"name":"Eugene van der Watt","@id":"https:\/\/dailyai.com\/#\/schema\/person\/7ce525c6d0c79838b7cc7cde96993cfa"},"headline":"Simply fine-tuning LLMs can remove alignment guardrails","datePublished":"2023-10-12T14:57:12+00:00","dateModified":"2023-10-13T10:04:42+00:00","mainEntityOfPage":{"@id":"https:\/\/dailyai.com\/2023\/10\/simply-fine-tuning-llms-can-remove-alignment-guardrails\/"},"wordCount":490,"publisher":{"@id":"https:\/\/dailyai.com\/#organization"},"image":{"@id":"https:\/\/dailyai.com\/2023\/10\/simply-fine-tuning-llms-can-remove-alignment-guardrails\/#primaryimage"},"thumbnailUrl":"https:\/\/dailyai.com\/wp-content\/uploads\/2023\/10\/unsafe-LLMs.jpg","keywords":["AI risks","ChatGPT","Llama 2","LLM","Meta","OpenAI"],"articleSection":["Industry"],"inLanguage":"pt-PT"},{"@type":"WebPage","@id":"https:\/\/dailyai.com\/2023\/10\/simply-fine-tuning-llms-can-remove-alignment-guardrails\/","url":"https:\/\/dailyai.com\/2023\/10\/simply-fine-tuning-llms-can-remove-alignment-guardrails\/","name":"O simples ajuste fino dos LLMs pode remover as barreiras de alinhamento | DailyAI","isPartOf":{"@id":"https:\/\/dailyai.com\/#website"},"primaryImageOfPage":{"@id":"https:\/\/dailyai.com\/2023\/10\/simply-fine-tuning-llms-can-remove-alignment-guardrails\/#primaryimage"},"image":{"@id":"https:\/\/dailyai.com\/2023\/10\/simply-fine-tuning-llms-can-remove-alignment-guardrails\/#primaryimage"},"thumbnailUrl":"https:\/\/dailyai.com\/wp-content\/uploads\/2023\/10\/unsafe-LLMs.jpg","datePublished":"2023-10-12T14:57:12+00:00","dateModified":"2023-10-13T10:04:42+00:00","breadcrumb":{"@id":"https:\/\/dailyai.com\/2023\/10\/simply-fine-tuning-llms-can-remove-alignment-guardrails\/#breadcrumb"},"inLanguage":"pt-PT","potentialAction":[{"@type":"ReadAction","target":["https:\/\/dailyai.com\/2023\/10\/simply-fine-tuning-llms-can-remove-alignment-guardrails\/"]}]},{"@type":"ImageObject","inLanguage":"pt-PT","@id":"https:\/\/dailyai.com\/2023\/10\/simply-fine-tuning-llms-can-remove-alignment-guardrails\/#primaryimage","url":"https:\/\/dailyai.com\/wp-content\/uploads\/2023\/10\/unsafe-LLMs.jpg","contentUrl":"https:\/\/dailyai.com\/wp-content\/uploads\/2023\/10\/unsafe-LLMs.jpg","width":1000,"height":667},{"@type":"BreadcrumbList","@id":"https:\/\/dailyai.com\/2023\/10\/simply-fine-tuning-llms-can-remove-alignment-guardrails\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/dailyai.com\/"},{"@type":"ListItem","position":2,"name":"Simply fine-tuning LLMs can remove alignment guardrails"}]},{"@type":"WebSite","@id":"https:\/\/dailyai.com\/#website","url":"https:\/\/dailyai.com\/","name":"DailyAI","description":"A sua dose di\u00e1ria de not\u00edcias sobre 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":"pt-PT"},{"@type":"Organization","@id":"https:\/\/dailyai.com\/#organization","name":"DailyAI","url":"https:\/\/dailyai.com\/","logo":{"@type":"ImageObject","inLanguage":"pt-PT","@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":"pt-PT","@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 vem de uma forma\u00e7\u00e3o em engenharia eletr\u00f3nica e adora tudo o que \u00e9 tecnologia. Quando faz uma pausa no consumo de not\u00edcias sobre IA, pode encontr\u00e1-lo \u00e0 mesa de snooker.","sameAs":["www.linkedin.com\/in\/eugene-van-der-watt-16828119"],"url":"https:\/\/dailyai.com\/pt\/author\/eugene\/"}]}},"_links":{"self":[{"href":"https:\/\/dailyai.com\/pt\/wp-json\/wp\/v2\/posts\/6384","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/dailyai.com\/pt\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/dailyai.com\/pt\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/dailyai.com\/pt\/wp-json\/wp\/v2\/users\/6"}],"replies":[{"embeddable":true,"href":"https:\/\/dailyai.com\/pt\/wp-json\/wp\/v2\/comments?post=6384"}],"version-history":[{"count":5,"href":"https:\/\/dailyai.com\/pt\/wp-json\/wp\/v2\/posts\/6384\/revisions"}],"predecessor-version":[{"id":6415,"href":"https:\/\/dailyai.com\/pt\/wp-json\/wp\/v2\/posts\/6384\/revisions\/6415"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/dailyai.com\/pt\/wp-json\/wp\/v2\/media\/6386"}],"wp:attachment":[{"href":"https:\/\/dailyai.com\/pt\/wp-json\/wp\/v2\/media?parent=6384"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/dailyai.com\/pt\/wp-json\/wp\/v2\/categories?post=6384"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/dailyai.com\/pt\/wp-json\/wp\/v2\/tags?post=6384"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}