{"id":180970,"date":"2023-10-25T23:20:59","date_gmt":"2023-10-25T17:50:59","guid":{"rendered":"https:\/\/dripp.zone\/news\/?p=180970"},"modified":"2023-10-25T23:20:59","modified_gmt":"2023-10-25T17:50:59","slug":"researchers-in-china-developed-a-hallucination-correction-engine-for-ai-models-crypto-news","status":"publish","type":"post","link":"https:\/\/dripp.zone\/news\/researchers-in-china-developed-a-hallucination-correction-engine-for-ai-models-crypto-news\/","title":{"rendered":"Researchers in China developed a hallucination correction engine for AI models &#8211; Crypto News"},"content":{"rendered":"<p><\/p>\n<div data-v-3d3be88a=\"\">\n<p>A team of scientists from the University of Science and Technology of China and Tencent\u2019s YouTu Lab have developed a tool to combat \u201challucination\u201d by artificial intelligence (AI) models.\u00a0<\/p>\n<p>Hallucination is the tendency for an AI model to generate outputs with a high level of confidence that don\u2019t appear based on information present in its training data. This problem <a rel=\"nofollow noopener\" target=\"_blank\" href=\"https:\/\/www.nytimes.com\/2023\/05\/01\/business\/ai-chatbots-hallucination.html\">permeates<\/a> large language model (LLM) research. Its effects can be seen in models such as OpenAI\u2019s ChatGPT and Anthropic\u2019s Claude. <\/p>\n<p><iframe loading=\"lazy\" title=\"[short] Woodpecker: Hallucination Correction for Multimodal Large Language Models\" width=\"740\" height=\"555\" src=\"https:\/\/www.youtube.com\/embed\/gz6v2eOSZTE?feature=oembed\" frameborder=\"0\" allow=\"accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share\" allowfullscreen><\/iframe><\/p>\n<p>The USTC\/Tencent team developed a tool called \u201cWoodpecker\u201d that they claim is capable of correcting hallucinations in multi-modal large language models (MLLMs). <\/p>\n<p>This subset of AI involves models such as GPT-4 (especially its visual variant, GPT-4V) and other systems that roll vision and\/or other processing into the generative AI modality alongside text-based language modelling. <\/p>\n<p>According to the team\u2019s pre-print research paper, Woodpecker <a rel=\"nofollow noopener\" target=\"_blank\" href=\"https:\/\/arxiv.org\/pdf\/2310.16045.pdf\">uses<\/a> three separate AI models, apart from the MLLM being corrected for hallucinations, to perform hallucination correction. <\/p>\n<p>These include GPT-3.5 turbo, Grounding DINO, and BLIP-2-FlanT5. Together, these models work as evaluators to identify hallucinations and instruct the model being corrected to re-generate its output in accordance with its data. <\/p>\n<figure><figcaption style=\"text-align: center\">In each of the above examples, an LLM hallucinates an incorrect answer (green background) to prompting (blue background). The corrected \u201cWoodpecker\u201d responses are shown with a red background. (Image source: <a rel=\"nofollow noopener\" target=\"_blank\" href=\"https:\/\/arxiv.org\/pdf\/2310.16045.pdf\">Yin, et. al., 2023<\/a>).<\/figcaption><\/figure>\n<p>To correct hallucinations, the AI models powering \u201cWoodpecker\u201d use a five-stage process that involves \u201ckey concept extraction, question formulation, visual knowledge validation, visual claim generation, and hallucination correction.\u201d <\/p>\n<p>The researchers claim these techniques provide additional transparency and \u201ca\u00a030.66%\/24.33% improvement in accuracy over the baseline MiniGPT-4\/mPLUG-Owl.\u201d They evaluated numerous \u201coff the shelf\u201d MLLMs using their method and concluded that Woodpecker could be \u201ceasily integrated into other MLLMs.\u201d<\/p>\n<p><strong><em>Related: <\/em><\/strong><a rel=\"nofollow noopener\" target=\"_blank\" href=\"https:\/\/cointelegraph.com\/news\/humans-ai-prefer-sycophantic-chatbot-answers-truth-study\" data-amp=\"https:\/\/cointelegraph-com.cdn.ampproject.org\/c\/s\/cointelegraph.com\/news\/humans-ai-prefer-sycophantic-chatbot-answers-truth-study\/amp\"><strong><em>Humans and AI often prefer sycophantic chatbot answers to the truth \u2014 Study<\/em><\/strong><\/a><\/p>\n<p>An evaluation version of Woodpecker is <a rel=\"nofollow noopener\" target=\"_blank\" href=\"https:\/\/60d1b7c6f5408b81d1.gradio.live\/\">available<\/a> on Gradio Live where anyone curious can check out the tool in action.<\/p>\n<\/div>\n","protected":false},"excerpt":{"rendered":"<p>A team of scientists from the University of Science and Technology of China and Tencent\u2019s YouTu Lab have developed a tool to combat \u201challucination\u201d by artificial intelligence (AI) models.\u00a0 Hallucination is the tendency for an AI model to generate outputs with a high level of confidence that don\u2019t appear based on information present in its [&hellip;]<\/p>\n","protected":false},"author":2,"featured_media":180971,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[4],"tags":[235,203,210,234,231,232,237,238,236,233],"class_list":["post-180970","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-blockchain","tag-bitcoin","tag-crypto-currency","tag-elon-musk","tag-ethereum","tag-hyperledger","tag-ibm","tag-mining","tag-nodes","tag-spacex","tag-tesla"],"_links":{"self":[{"href":"https:\/\/dripp.zone\/news\/wp-json\/wp\/v2\/posts\/180970","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/dripp.zone\/news\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/dripp.zone\/news\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/dripp.zone\/news\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/dripp.zone\/news\/wp-json\/wp\/v2\/comments?post=180970"}],"version-history":[{"count":2,"href":"https:\/\/dripp.zone\/news\/wp-json\/wp\/v2\/posts\/180970\/revisions"}],"predecessor-version":[{"id":180973,"href":"https:\/\/dripp.zone\/news\/wp-json\/wp\/v2\/posts\/180970\/revisions\/180973"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/dripp.zone\/news\/wp-json\/wp\/v2\/media\/180971"}],"wp:attachment":[{"href":"https:\/\/dripp.zone\/news\/wp-json\/wp\/v2\/media?parent=180970"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/dripp.zone\/news\/wp-json\/wp\/v2\/categories?post=180970"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/dripp.zone\/news\/wp-json\/wp\/v2\/tags?post=180970"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}