{"id":11863,"date":"2024-08-17T19:26:30","date_gmt":"2024-08-18T02:26:30","guid":{"rendered":"https:\/\/mattfife.com\/?p=11863"},"modified":"2025-05-02T21:44:29","modified_gmt":"2025-05-03T04:44:29","slug":"training-on-your-generated-output-destroys-generative-ai-models","status":"publish","type":"post","link":"https:\/\/mattfife.com\/?p=11863","title":{"rendered":"Training an AI model on its own generated output destroys the model"},"content":{"rendered":"\n<p>Describing a situation much like the dangers of genetic inbreeding, computer scientists Matyas Bohacek and Hany Farid wrote a paper that describes how AI image generators that start training with their own generated data quickly start deteriorating.<\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter size-large is-resized\"><img data-recalc-dims=\"1\" loading=\"lazy\" decoding=\"async\" width=\"640\" height=\"683\" data-attachment-id=\"11865\" data-permalink=\"https:\/\/mattfife.com\/?attachment_id=11865\" data-orig-file=\"https:\/\/i0.wp.com\/mattfife.com\/wp-content\/themes\/mattTheme\/headerimgs\/2024\/08\/image-3.png?fit=1297%2C1383&amp;ssl=1\" data-orig-size=\"1297,1383\" data-comments-opened=\"1\" data-image-meta=\"{&quot;aperture&quot;:&quot;0&quot;,&quot;credit&quot;:&quot;&quot;,&quot;camera&quot;:&quot;&quot;,&quot;caption&quot;:&quot;&quot;,&quot;created_timestamp&quot;:&quot;0&quot;,&quot;copyright&quot;:&quot;&quot;,&quot;focal_length&quot;:&quot;0&quot;,&quot;iso&quot;:&quot;0&quot;,&quot;shutter_speed&quot;:&quot;0&quot;,&quot;title&quot;:&quot;&quot;,&quot;orientation&quot;:&quot;0&quot;}\" data-image-title=\"image\" data-image-description=\"\" data-image-caption=\"\" data-large-file=\"https:\/\/i0.wp.com\/mattfife.com\/wp-content\/themes\/mattTheme\/headerimgs\/2024\/08\/image-3.png?fit=640%2C683&amp;ssl=1\" src=\"https:\/\/i0.wp.com\/mattfife.com\/wp-content\/themes\/mattTheme\/headerimgs\/2024\/08\/image-3.png?resize=640%2C683&#038;ssl=1\" alt=\"\" class=\"wp-image-11865\" style=\"width:726px;height:auto\" srcset=\"https:\/\/i0.wp.com\/mattfife.com\/wp-content\/themes\/mattTheme\/headerimgs\/2024\/08\/image-3.png?resize=960%2C1024&amp;ssl=1 960w, https:\/\/i0.wp.com\/mattfife.com\/wp-content\/themes\/mattTheme\/headerimgs\/2024\/08\/image-3.png?resize=281%2C300&amp;ssl=1 281w, https:\/\/i0.wp.com\/mattfife.com\/wp-content\/themes\/mattTheme\/headerimgs\/2024\/08\/image-3.png?resize=768%2C819&amp;ssl=1 768w, https:\/\/i0.wp.com\/mattfife.com\/wp-content\/themes\/mattTheme\/headerimgs\/2024\/08\/image-3.png?resize=253%2C270&amp;ssl=1 253w, https:\/\/i0.wp.com\/mattfife.com\/wp-content\/themes\/mattTheme\/headerimgs\/2024\/08\/image-3.png?w=1297&amp;ssl=1 1297w\" sizes=\"auto, (max-width: 640px) 100vw, 640px\" \/><\/figure>\n<\/div>\n\n\n<p>&#8216;<a href=\"https:\/\/ar5iv.labs.arxiv.org\/html\/2311.12202\" data-type=\"link\" data-id=\"https:\/\/ar5iv.labs.arxiv.org\/html\/2311.12202\">Nepotistically Trained Generative-AI Models Collapse<\/a>&#8216; shows that training an AI image generator on AI generated images quickly leads to a deterioration in the quality of output which can only be fixed by re-introducing real images. <\/p>\n\n\n\n<p>In\u00a0<a href=\"https:\/\/go.redirectingat.com\/?id=92X1519156&amp;xcust=cbq_gb_1139793430362001478&amp;xs=1&amp;url=https%3A%2F%2Fwww.nature.com%2Farticles%2Fd41586-024-02355-z&amp;sref=https%3A%2F%2Fwww.creativebloq.com%2Fai%2Fai-art%2Fresearch-shows-ai-image-generators-could-be-their-own-demise\" target=\"_blank\" rel=\"noreferrer noopener\">Nature<\/a>, a more recent study found a similar effect in text generation, with the use of synthetic data leading to increasingly nonsensical results.<\/p>\n\n\n\n<p>Artifactory:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><a href=\"https:\/\/arxiv.org\/abs\/2311.12202\">https:\/\/arxiv.org\/abs\/2311.12202<\/a><\/li>\n\n\n\n<li><a href=\"https:\/\/www.creativebloq.com\/ai\/ai-art\/research-shows-ai-image-generators-could-be-their-own-demise\">https:\/\/www.creativebloq.com\/ai\/ai-art\/research-shows-ai-image-generators-could-be-their-own-demise<\/a><\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"<p>Describing a situation much like the dangers of genetic inbreeding, computer scientists Matyas Bohacek and Hany Farid wrote a paper that describes how AI image generators that start training with their own generated data quickly start deteriorating. &#8216;Nepotistically Trained Generative-AI Models Collapse&#8216; shows that training an AI image generator on AI generated images quickly leads to a deterioration in the quality of output which can only be fixed by re-introducing real images. In\u00a0Nature, a more recent study found a similar&#8230;<\/p>\n<p class=\"read-more\"><a class=\"btn btn-default\" href=\"https:\/\/mattfife.com\/?p=11863\"> Read More<span class=\"screen-reader-text\">  Read More<\/span><\/a><\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"jetpack_post_was_ever_published":false,"_jetpack_newsletter_access":"","_jetpack_dont_email_post_to_subs":false,"_jetpack_newsletter_tier_id":0,"_jetpack_memberships_contains_paywalled_content":false,"_jetpack_memberships_contains_paid_content":false,"footnotes":"","jetpack_publicize_message":"","jetpack_publicize_feature_enabled":true,"jetpack_social_post_already_shared":true,"jetpack_social_options":{"image_generator_settings":{"template":"highway","default_image_id":0,"font":"","enabled":false},"version":2}},"categories":[28,9],"tags":[],"class_list":["post-11863","post","type-post","status-publish","format-standard","hentry","category-ai","category-cool"],"jetpack_publicize_connections":[],"jetpack_featured_media_url":"","jetpack_sharing_enabled":true,"jetpack_shortlink":"https:\/\/wp.me\/p4WECr-35l","jetpack-related-posts":[],"_links":{"self":[{"href":"https:\/\/mattfife.com\/index.php?rest_route=\/wp\/v2\/posts\/11863","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/mattfife.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/mattfife.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/mattfife.com\/index.php?rest_route=\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/mattfife.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=11863"}],"version-history":[{"count":6,"href":"https:\/\/mattfife.com\/index.php?rest_route=\/wp\/v2\/posts\/11863\/revisions"}],"predecessor-version":[{"id":13952,"href":"https:\/\/mattfife.com\/index.php?rest_route=\/wp\/v2\/posts\/11863\/revisions\/13952"}],"wp:attachment":[{"href":"https:\/\/mattfife.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=11863"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/mattfife.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=11863"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/mattfife.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=11863"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}