{"id":134242,"date":"2024-10-01T09:05:05","date_gmt":"2024-10-01T06:05:05","guid":{"rendered":"https:\/\/tietopankki.crnet.fi\/fi\/news\/new-ai-models-are-more-likely-to-give-a-wrong-answer-than-admit-they-dont-know\/"},"modified":"2024-10-01T09:05:05","modified_gmt":"2024-10-01T06:05:05","slug":"new-ai-models-are-more-likely-to-give-a-wrong-answer-than-admit-they-dont-know","status":"publish","type":"post","link":"https:\/\/tietopankki.crnet.fi\/fi\/news\/new-ai-models-are-more-likely-to-give-a-wrong-answer-than-admit-they-dont-know\/","title":{"rendered":"New AI models are more likely to give a wrong answer than admit they don\u2019t know"},"content":{"rendered":"<div class=\"mp_wrapper\">\n  <div class=\"mepr-unauthorized-message\">\n    <p>You are unauthorized to view this page.<\/p>\n  <\/div>\n  <div class=\"mepr-login-form-wrap\">\n            \n<div class=\"mp_wrapper mp_login_form\">\n                  <!-- mp-login-form-start -->     <form name=\"mepr_loginform\" id=\"mepr_loginform\" class=\"mepr-form\" action=\"https:\/\/tietopankki.crnet.fi\/fi\/login-2\/\" method=\"post\">\n            <div class=\"mp-form-row mepr_username\">\n        <div class=\"mp-form-label\">\n                              <label for=\"user_login\">Username<\/label>\n        <\/div>\n        <input type=\"text\" name=\"log\" id=\"user_login\" value=\"\" \/>\n      <\/div>\n      <div class=\"mp-form-row mepr_password\">\n        <div class=\"mp-form-label\">\n          <label for=\"user_pass\">Password<\/label>\n          <div class=\"mp-hide-pw\">\n            <input type=\"password\" name=\"pwd\" id=\"user_pass\" value=\"\" \/>\n            <button type=\"button\" class=\"button mp-hide-pw hide-if-no-js\" data-toggle=\"0\" aria-label=\"Show password\">\n              <span class=\"dashicons dashicons-visibility\" aria-hidden=\"true\"><\/span>\n            <\/button>\n          <\/div>\n        <\/div>\n      <\/div>\n            <div>\n        <label><input name=\"rememberme\" type=\"checkbox\" id=\"rememberme\" value=\"forever\" \/> Remember Me<\/label>\n      <\/div>\n      <div class=\"mp-spacer\">&nbsp;<\/div>\n      <div class=\"submit\">\n        <input type=\"submit\" name=\"wp-submit\" id=\"wp-submit\" class=\"button-primary mepr-share-button \" value=\"Log In\" \/>\n        <input type=\"hidden\" name=\"redirect_to\" value=\"\/fi\/wp-json\/wp\/v2\/posts\/134242\" \/>\n        <input type=\"hidden\" name=\"mepr_process_login_form\" value=\"true\" \/>\n        <input type=\"hidden\" name=\"mepr_is_login_page\" value=\"false\" \/>\n      <\/div>\n    <\/form>\n    <div class=\"mp-spacer\">&nbsp;<\/div>\n    <div class=\"mepr-login-actions\">\n        <a\n          href=\"https:\/\/tietopankki.crnet.fi\/fi\/login-2\/?action=forgot_password\"\n          title=\"Click here to reset your password\"\n        >\n          Forgot Password        <\/a>\n    <\/div>\n\n      \n    <!-- mp-login-form-end --> \n  <\/div>\n      <\/div>\n<\/div>\n","protected":false},"excerpt":{"rendered":"<div class=\"smart_content_wrapper\">\n<p>The more scaled up LLMs get, the more likely they are to fudge an answer rather than admit their ignorance. According to a new study, the more advanced an AI large language model (LLM) becomes, the less likely it is to admit it can&#8217;t answer a query. Newer large language models (LLMs) are less likely to admit they don\u2019t know an answer to a user\u2019s question making them less reliable, according to a new study.<\/p>\n<p>Artificial intelligence (AI) researchers from the Universitat Polit\u00e8cnica de Val\u00e8ncia in Spain tested the latest versions of BigScience\u2019s BLOOM, Meta\u2019s Llama, and OpenAI&#8217;s GPT for accuracy by asking each model thousands of questions on maths, science, and geography. Researchers compared the quality of the answers of each model and classified them into correct, incorrect, or avoidant answers.<\/p>\n<p>Microsoft claims its new AI correction feature can fix hallucinations. Does it work?<br \/>\nThe study, which was published in the journal Nature, found that accuracy on more challenging problems improved with each new model. Still, they tended to be less transparent about whether they could answer a question correctly. The earlier LLM models would say they could not find the answers or needed more information to come to an answer, but new models were more likely to guess and produce incorrect responses even to easy questions.<\/p>\n<p>LLMs are deep learning algorithms that use AI to understand, predict, and generate new content based on data sets. While the new models could solve more complex problems with more accuracy, the LLMs in the study still made some mistakes when answering basic questions. &#8220;Full reliability is not even achieved at very low difficulty levels,&#8221; according to the research paper. &#8220;Although the models can solve highly challenging instances, they also still fail at very simple ones&#8221;.<\/p>\n<p>Prisoners in Finland are being employed as data labellers to improve accuracy of AI models. This is the case with OpenAI\u2019s GPT-4, where the number of &#8220;avoidant&#8221; answers significantly dropped off from its previous model, GPT-3.5. \u201cThis does not match the expectation that more recent LLMs would more successfully avoid answering outside their operating range,\u201d the study authors said. Researchers concluded then that there&#8217;s &#8220;no apparent improvement&#8221; for the models even though the technology has been scaled up.<\/p>\n<p>Source: <a rel=\"noopener\" href=\"https:\/\/www.euronews.com\/next\/2024\/10\/01\/new-ai-models-are-more-likely-to-give-a-wrong-answer-than-admit-they-dont-know\">Euronews<\/a><\/p>\n<\/div>\n<p>The post <a rel=\"noopener\" href=\"https:\/\/vastuullisuusuutiset.fi\/fi\/vaufien\/new-ai-models-are-more-likely-to-give-a-wrong-answer-than-admit-they-dont-know\/\">New AI models are more likely to give a wrong answer than admit they don\u2019t know<\/a> appeared first on <a rel=\"noopener\" href=\"https:\/\/vastuullisuusuutiset.fi\/fi\">Vastuullisuusuutiset.fi<\/a>.<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_jetpack_memberships_contains_paid_content":false,"footnotes":""},"categories":[22],"tags":[16313,606,607,608],"class_list":["post-134242","post","type-post","status-publish","format-standard","hentry","category-news","tag-new-ai-models","tag-vau_viikkokatsaus","tag-vauen","tag-vaufien"],"jetpack_featured_media_url":"","jetpack_sharing_enabled":true,"_links":{"self":[{"href":"https:\/\/tietopankki.crnet.fi\/fi\/wp-json\/wp\/v2\/posts\/134242","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/tietopankki.crnet.fi\/fi\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/tietopankki.crnet.fi\/fi\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/tietopankki.crnet.fi\/fi\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/tietopankki.crnet.fi\/fi\/wp-json\/wp\/v2\/comments?post=134242"}],"version-history":[{"count":0,"href":"https:\/\/tietopankki.crnet.fi\/fi\/wp-json\/wp\/v2\/posts\/134242\/revisions"}],"wp:attachment":[{"href":"https:\/\/tietopankki.crnet.fi\/fi\/wp-json\/wp\/v2\/media?parent=134242"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/tietopankki.crnet.fi\/fi\/wp-json\/wp\/v2\/categories?post=134242"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/tietopankki.crnet.fi\/fi\/wp-json\/wp\/v2\/tags?post=134242"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}