{"id":381431,"date":"2025-02-10T16:48:11","date_gmt":"2025-02-10T16:48:11","guid":{"rendered":"https:\/\/www.techopedia.com\/?p=381431"},"modified":"2025-02-10T16:48:11","modified_gmt":"2025-02-10T16:48:11","slug":"diffbot-ai-model-knowledge-graph-technology","status":"publish","type":"post","link":"https:\/\/www.techopedia.com\/diffbot-ai-model-knowledge-graph-technology","title":{"rendered":"Diffbot\u2019s AI Model Suggests &#8220;Smaller Is Better&#8221; for LLMs"},"content":{"rendered":"<p><a href=\"https:\/\/www.techopedia.com\/definition\/190\/artificial-intelligence-ai\">Artificial intelligence<\/a> (AI) is nothing without data. Hence, the underlying assumption has been that whoever wins the AI race must have <a href=\"https:\/\/www.techopedia.com\/ai-training-data-has-run-out-what-is-next\">a pool of training data<\/a> that&#8217;s not just vast but better than others.<\/p>\n<p>Take <a href=\"https:\/\/www.techopedia.com\/is-gpt4-worth-paying-for\">OpenAI&#8217;s GPT-4<\/a> and Meta&#8217;s Llama 3.1 as examples; they run on 1.7 trillion and 405 billion parameters, respectively, and have gained popularity due to their gigantic training datasets.<\/p>\n<p>Each newly released <a href=\"https:\/\/www.techopedia.com\/definition\/34948\/large-language-model-llm\">large language model<\/a> (LLM) often shatters the parameter record of its predecessor, a trend that has become a central benchmark for measuring progress in LLM development on <a href=\"https:\/\/www.techopedia.com\/definition\/hugging-face\">Hugging Face<\/a>.<\/p>\n<p>But is bigger always better? Diffbot Technologies, a California-based startup known for its knowledge graph technology, does not think so.<\/p>\n<div class=\"su-note\"  style=\"border-color:#e5e54c;border-radius:3px;-moz-border-radius:3px;-webkit-border-radius:3px;\"><div class=\"su-note-inner su-u-clearfix su-u-trim\" style=\"background-color:#FFFF66;border-color:#ffffff;color:#333333;border-radius:3px;-moz-border-radius:3px;-webkit-border-radius:3px;\">\n<h2><span id=\"key_takeaways\">Key Takeaways<\/span><\/h2>\n<ul>\n<li>Diffbot&#8217;s AI model challenges larger LLMs by emphasizing factual accuracy over sheer parameter size.<\/li>\n<li>It uses a proprietary knowledge graph with real-time updates instead of relying on pre-trained data.<\/li>\n<li>Diffbot&#8217;s graph retrieval-augmented generation (GraphRAG) allows dynamic knowledge retrieval, reducing reliance on static datasets.<\/li>\n<li>Benchmark tests suggest Diffbot outperforms leading LLMs like GPT-4 and Gemini in real-time factual accuracy.<\/li>\n<li>Experts believe Diffbot&#8217;s hybrid approach offers a step toward solving AI hallucinations but acknowledge that no model is entirely immune.<\/li>\n<\/ul>\n<\/div><\/div>\n<div>\n    <section class=\"toc-sticky w-100 bg-white \">\n        <div class=\"toc-sticky__container container\">\n            <div class=\"toc-sticky__open d-flex align-items-end\" data-bs-toggle=\"collapse\"\n                 aria-controls=\"multiCollapse1\"\n                 data-bs-target=\"#multiCollapse1\">\n                <button\n                        class=\"btn btn-primary collapse-action-btn p-1 rounded-circle\"\n                        type=\"button\"\n                >\n                    <i class=\"icon-chevron-up\"><\/i>\n                <\/button>\n                <span id=\"toc-current-title\" class=\"d-flex align-items-center\">\n                    Table of Contents\n                <\/span>\n                <span class=\"toc-main-title-permanent\">Table of Contents<\/span>\n            <\/div>\n            <div\n                    class=\"collapse my-3\"\n                    id=\"multiCollapse1\"\n            >\n                <ol class=\"StepProgress\">\n                                                                        <li class=\"StepProgress-item current\">\n                                                <div class=\"StepProgress-item__group\">\n                            <a data-id=\"key_takeaways\"\n                               onclick=handleScrollToTitle(\"key_takeaways\")\n                               class=\"StepProgress-item__link\" data-level=\"2\">Key Takeaways<\/a>\n                                                    <\/div>\n                                                <\/li>\n                                                                        <li class=\"StepProgress-item\">\n                                                <div class=\"StepProgress-item__group\">\n                            <a data-id=\"diffbots_different_look_at_ai_models\"\n                               onclick=handleScrollToTitle(\"diffbots_different_look_at_ai_models\")\n                               class=\"StepProgress-item__link\" data-level=\"2\">Diffbot's Different Look at AI Models<\/a>\n                                                    <\/div>\n                                                <\/li>\n                                                                        <li class=\"StepProgress-item\">\n                                                <div class=\"StepProgress-item__group\">\n                            <a data-id=\"our_quick_test_of_diffbots_ai_model_on_slippery_questions\"\n                               onclick=handleScrollToTitle(\"our_quick_test_of_diffbots_ai_model_on_slippery_questions\")\n                               class=\"StepProgress-item__link\" data-level=\"2\">Our Quick Test of Diffbot's AI Model on Slippery Questions<\/a>\n                                                    <\/div>\n                                                <\/li>\n                                                                        <li class=\"StepProgress-item\">\n                                                <div class=\"StepProgress-item__group\">\n                            <a data-id=\"can_diffbots_ai_model_solve_the_ai_hallucination_puzzle\"\n                               onclick=handleScrollToTitle(\"can_diffbots_ai_model_solve_the_ai_hallucination_puzzle\")\n                               class=\"StepProgress-item__link\" data-level=\"2\">Can Diffbot's AI Model Solve the AI Hallucination Puzzle?<\/a>\n                                                    <\/div>\n                                                <\/li>\n                                                                        <li class=\"StepProgress-item\">\n                                                <div class=\"StepProgress-item__group\">\n                            <a data-id=\"the_bottom_line\"\n                               onclick=handleScrollToTitle(\"the_bottom_line\")\n                               class=\"StepProgress-item__link\" data-level=\"2\">The Bottom Line<\/a>\n                                                    <\/div>\n                                                <\/li>\n                                                                        <li class=\"StepProgress-item\">\n                                                <div class=\"StepProgress-item__group\">\n                            <a data-id=\"faqs\"\n                               onclick=handleScrollToTitle(\"faqs\")\n                               class=\"StepProgress-item__link\" data-level=\"2\">FAQs<\/a>\n                                                    <\/div>\n                                                <\/li>\n                                                                        <li class=\"StepProgress-item\">\n                                                <div class=\"StepProgress-item__group\">\n                            <a data-id=\"references\"\n                               onclick=handleScrollToTitle(\"references\")\n                               class=\"StepProgress-item__link\" data-level=\"2\">References<\/a>\n                                                    <\/div>\n                                                <\/li>\n                                    <\/ol>\n                <div class=\"toc-sticky__container__disperse\"><\/div>\n                <div class=\"toc-sticky__btn-top justify-content-center\">\n                    <button class=\"btn btn-link d-flex align-items-center\" onclick={btnTop()}><p>Show Full Guide<\/p><img decoding=\"async\"\n                                src=\"\/wp-content\/plugins\/table-of-contents\/assets\/images\/arrow-up-circle.svg\"\n                                alt=\"arrow-up-circle\">\n                    <\/button>\n                <\/div>\n            <\/div>\n        <\/div>\n    <\/section>\n    <div class=\"toc-sticky-list\">\n        <div class=\"toc-sticky__container container\">\n            <div class=\"toc-sticky__open d-flex align-items-end\" data-bs-toggle=\"collapse\" aria-controls=\"multiCollapse2\" data-bs-target=\"#multiCollapse2\">\n                <button class=\"btn btn-primary collapse-action-btn p-1 rounded-circle\"\n                        type=\"button\">\n                    <i class=\"icon-chevron-up up\"><\/i>\n                <\/button>\n                <span id=\"toc-current-title\" class=\"d-flex align-items-center\">\n                    Table of Contents\n                <\/span>\n            <\/div>\n            <div  class=\"collapse my-3\" id=\"multiCollapse2\">\n                <ol class=\"StepProgress\">\n                    \t\t\t\t\t \t\t\t\t\t                                                     <li class=\"StepProgress-item current \">\n                                                    <div class=\"StepProgress-item__group\">\n                                <a data-id=\"key_takeaways\"\n                                   onclick=handleScrollToTitle(\"key_takeaways\")\n                                   class=\"StepProgress-item__link\"\n                                   data-level=\"2\">Key Takeaways<\/a>\n                                                            <\/div>\n                                                    <\/li>\n                    \t\t\t\t\t \t\t\t\t\t                                                     <li class=\"StepProgress-item \">\n                                                    <div class=\"StepProgress-item__group\">\n                                <a data-id=\"diffbots_different_look_at_ai_models\"\n                                   onclick=handleScrollToTitle(\"diffbots_different_look_at_ai_models\")\n                                   class=\"StepProgress-item__link\"\n                                   data-level=\"2\">Diffbot's Different Look at AI Models<\/a>\n                                                            <\/div>\n                                                    <\/li>\n                    \t\t\t\t\t \t\t\t\t\t                                                     <li class=\"StepProgress-item \">\n                                                    <div class=\"StepProgress-item__group\">\n                                <a data-id=\"our_quick_test_of_diffbots_ai_model_on_slippery_questions\"\n                                   onclick=handleScrollToTitle(\"our_quick_test_of_diffbots_ai_model_on_slippery_questions\")\n                                   class=\"StepProgress-item__link\"\n                                   data-level=\"2\">Our Quick Test of Diffbot's AI Model on Slippery Questions<\/a>\n                                                            <\/div>\n                                                    <\/li>\n                    \t\t\t\t\t \t\t\t\t\t\n\t\t\t\t\t\t <span class=\"show_moretoc\">Show Full Guide<\/span>\n\t\t\t\t\t\t \t\t\t\t\t \t\t\t\t\t\t  \t\t\t\t\t                                                      <li class=\"StepProgress-item ms_hidetoc\">\n                                                    <div class=\"StepProgress-item__group\">\n                                <a data-id=\"can_diffbots_ai_model_solve_the_ai_hallucination_puzzle\"\n                                   onclick=handleScrollToTitle(\"can_diffbots_ai_model_solve_the_ai_hallucination_puzzle\")\n                                   class=\"StepProgress-item__link\"\n                                   data-level=\"2\">Can Diffbot's AI Model Solve the AI Hallucination Puzzle?<\/a>\n                                                            <\/div>\n                                                    <\/li>\n                    \t\t\t\t\t \t\t\t\t\t \t\t\t\t\t\t  \t\t\t\t\t                                                      <li class=\"StepProgress-item ms_hidetoc\">\n                                                    <div class=\"StepProgress-item__group\">\n                                <a data-id=\"the_bottom_line\"\n                                   onclick=handleScrollToTitle(\"the_bottom_line\")\n                                   class=\"StepProgress-item__link\"\n                                   data-level=\"2\">The Bottom Line<\/a>\n                                                            <\/div>\n                                                    <\/li>\n                    \t\t\t\t\t \t\t\t\t\t \t\t\t\t\t\t  \t\t\t\t\t                                                      <li class=\"StepProgress-item ms_hidetoc\">\n                                                    <div class=\"StepProgress-item__group\">\n                                <a data-id=\"faqs\"\n                                   onclick=handleScrollToTitle(\"faqs\")\n                                   class=\"StepProgress-item__link\"\n                                   data-level=\"2\">FAQs<\/a>\n                                                            <\/div>\n                                                    <\/li>\n                    \t\t\t\t\t \t\t\t\t\t \t\t\t\t\t\t  \t\t\t\t\t                                                      <li class=\"StepProgress-item ms_hidetoc\">\n                                                    <div class=\"StepProgress-item__group\">\n                                <a data-id=\"references\"\n                                   onclick=handleScrollToTitle(\"references\")\n                                   class=\"StepProgress-item__link\"\n                                   data-level=\"2\">References<\/a>\n                                                            <\/div>\n                                                    <\/li>\n                                    <\/ol>\n            <\/div>\n            <div class=\"toc-sticky__container__disperse\"><\/div>\n        <\/div>\n    <\/div>\n<\/div>\n\n<script id=\"toc-js\">\n    window.addEventListener(\"DOMContentLoaded\", () => {\n        const header = document.querySelector(\".header_wrapper\");\n\n        const pageLegend = document.querySelector('#multiCollapse1');\n        const pageLegendList = document.querySelector('#multiCollapse2');\n        const pageLegendCollapse = new bootstrap.Collapse(pageLegend, {toggle: document.querySelector(\".toc-sticky\").classList.contains('sticky')});\n\n        \/**\n         * Changing current title\n         *\/\n        (function (pageLegend) {\n            const titleNodes = pageLegend.querySelectorAll('.StepProgress-item__link');\n\n            if (!titleNodes.length) return;\n\n            const titles = [...titleNodes].map((itm, i) => ({\n                id: itm.getAttribute('data-id'),\n                text: itm.textContent,\n                level: itm.getAttribute('data-level'),\n                linkNode: itm,\n                titleNode: document.getElementById(itm.getAttribute('data-id')),\n                index: i,\n            }));\n\n            \/**\n             * Source: https:\/\/www.sitepoint.com\/throttle-scroll-events\/\n             * @param {Function} fn\n             * @param {number} wait\n             * @returns {(function(): void)|*}\n             *\/\n            const throttle = (fn, wait) => {\n                let time = Date.now();\n                return function () {\n                    if ((time + wait - Date.now()) < 0) {\n                        fn();\n                        time = Date.now();\n                    }\n                }\n            }\n\n            const changeCurrentTitle = () => {\n                const documentScrollTop = window.pageYOffset || document.documentElement.scrollTop || document.body.scrollTop || 0;\n                let current = 0;\n\n                \/\/ Title\n                titles.forEach((itm, i) => {\n\t\t\t\t\/\/console.log(itm)\n                    const itmOffsetTop = itm.titleNode ? itm.titleNode.offsetTop - 100 : 0;\n\n                    if (documentScrollTop >= itmOffsetTop) {\n                        document.getElementById('toc-current-title').innerHTML = itm.text;\n                        document.getElementById('toc-current-title').setAttribute('data-current-id', itm.id);\n                        document.getElementById('toc-current-title').setAttribute('data-current-level', itm.level);\n                        current = i;\n                    }\n                })\n\n                \/\/ close all list and open sub list if needed\n                if (document.querySelector(\".toc-sticky\").classList.contains('sticky')) {\n                    document.querySelectorAll('.subList-in-progress').forEach((el) => {\n                        el.children[1].classList.remove('show');\n                        el.getElementsByClassName('icon-chevron-down')[0].classList.remove('up');\n                    });\n                    const currentEl = titles[current];\n                    currentEl.linkNode.classList.add('show');\n                }\n\n                titles.forEach((itm, i) => {\n                    itm.linkNode.parentNode.parentNode.classList.remove('current', 'is-done');\n                    if (current > i) {\n                        itm.linkNode.parentNode.parentNode.classList.add('is-done')\n                    };\n                    if (current === i) {\n                      itm.linkNode.parentNode.parentNode.classList.add('current');\n                    };\n                })\n\n            }\n\n            changeCurrentTitle();\n\n            document.addEventListener('scroll', throttle(changeCurrentTitle, 50));\n        })(pageLegend);\n\n\n        \/**\n         *  Collapse\n         *\/\n        (function (pageLegend, header) {\n            const icon = pageLegend.parentNode.querySelector(\".collapse-action-btn i\");\n\n            const collapseToggle = (status) => (e) => {\n                if (!e.target.isEqualNode(pageLegend)) return;\n\n                icon.classList.toggle(\"up\");\n\n                const containerHeight = pageLegend.getBoundingClientRect().height;\n\n                const showSubtitleContent = () => {\n                    const currentId = document.getElementById('toc-current-title').getAttribute('data-current-id');\n                    const currentLevel = document.getElementById('toc-current-title').getAttribute('data-current-level');\n                    const currentSubTitle = currentLevel == 3 ? document.querySelector(`a[data-id=\"${currentId}\"]`).parentNode.parentNode.parentNode : false;\n\n                    if (!currentSubTitle) return;\n                    new bootstrap.Collapse(currentSubTitle, {toggle: false}).show();\n                }\n\n                showSubtitleContent();\n                if (status === 'shown' && document.querySelector(\".toc-sticky\").classList.contains('sticky') ) {\n                    document.querySelector('html').classList.remove('overflow-hidden');\n                    pageLegend.classList.add('overflow-auto');\n                    pageLegend.style.height = `calc(100vh - ${header.getBoundingClientRect().height + document.querySelector('.toc-sticky__open').getBoundingClientRect().height + 16}px)`;\n                } else if (status === 'hide') {\n                    document.querySelector('html').classList.remove('overflow-hidden');\n                    pageLegend.classList.remove('overflow-auto');\n                    pageLegend.style.height = 'auto';\n                }\n            }\n\n            pageLegend.addEventListener('shown.bs.collapse', collapseToggle('shown'));\n            pageLegend.addEventListener('hide.bs.collapse', collapseToggle('hide'));\n        })(pageLegend, header);\n\n        \/**\n         * Collapse sub-titles\n         *\/\n        (function (pageLegend) {\n            const collapseEls = pageLegend.querySelectorAll('.collapse');\n\n            collapseEls.forEach(function (el) {\n\n                const toggleArrowDirection = function (e) {\n                    if (!e.target.isEqualNode(el)) return;\n\n                    const id = this.getAttribute('id');\n                    document.querySelector(`.collapse-action-btn[data-bs-target=\"#${id}\"] .icon-chevron-down`).classList.toggle('up');\n                }\n                el.addEventListener('shown.bs.collapse', toggleArrowDirection);\n                el.addEventListener('hide.bs.collapse', toggleArrowDirection);\n            })\n        })(pageLegend);\n\n        \/**\n         *  Collapse main title\n         *\/\n        (function (pageLegendList) {\n            const icon = pageLegendList.parentNode.querySelector(\".collapse-action-btn i\");\n\n            const collapseToggle = () => (e) => {\n                if (!e.target.isEqualNode(pageLegendList)) return;\n\n                icon.classList.toggle(\"up\");\n\n            }\n            pageLegendList.addEventListener('shown.bs.collapse', collapseToggle());\n            pageLegendList.addEventListener('hide.bs.collapse', collapseToggle());\n        })(pageLegendList);\n\n        (function (pageLegendList) {\n            const collapseEls = pageLegendList.querySelectorAll('.collapse');\n\n            collapseEls.forEach(function (el) {\n\n                const toggleArrowDirection = function (e) {\n                    if (!e.target.isEqualNode(el)) return;\n\n                    const id = this.getAttribute('id');\n                    document.querySelector(`.toc-sticky-list .collapse-action-btn[data-bs-target=\"#${id}\"] .icon-chevron-down`).classList.toggle('up');\n                }\n                el.addEventListener('shown.bs.collapse', toggleArrowDirection);\n                el.addEventListener('hide.bs.collapse', toggleArrowDirection);\n            })\n        })(pageLegendList);\n\n        \/**\n         * Sticky functionality\n         * Source: https:\/\/stackoverflow.com\/questions\/17893771\/javascript-sticky-div-after-scroll\n         *\/\n        (function (header, pageLegendCollapse) {\n            \/\/ set everything outside the onscroll event (less work per scroll)\n            const target = document.querySelector(\".toc-sticky\");\n            const targetListStatic = document.querySelector(\".toc-sticky-list\");\n\n            if (!target || !header) return;\n\n            const headerHeight = header.getBoundingClientRect().height;\n            const targetHeight = targetListStatic.getBoundingClientRect().height;\n\n            \/\/ -headerHeight so it won't be jumpy\n            const stop = targetListStatic.offsetTop + headerHeight + targetHeight;\n            const docBody =\n                document.documentElement || document.body.parentNode || document.body;\n            const hasOffset = window.pageYOffset !== undefined;\n\n            const applySticky = function () {\n                \/\/ cross-browser compatible scrollTop.\n                const scrollTop = hasOffset ? window.pageYOffset : docBody.scrollTop;\n\n                \/\/ if user scrolls to headerHeight from the top of the target div\n                if (scrollTop >= stop) {\n                    pageLegendCollapse.hide();\n                    \/\/ stick the div\n                    target.classList.add(\"sticky\");\n                    \/\/target.style.marginTop = `${headerHeight}px`;\n                } else {\n                    pageLegendCollapse.show();\n                    \/\/ release the div\n                    target.classList.remove(\"sticky\");\n                    target.style.marginTop = \"\";\n                }\n            }\n\n            applySticky();\n\n            window.addEventListener('scroll', applySticky);\n        })(header, pageLegendCollapse);\n\tjQuery('span.show_moretoc').click(function(){\n\t\tjQuery('span.show_moretoc').hide();\n\t   jQuery('.ms_hidetoc').show();\n\t});\n    });\n\n    \/\/ When the user clicks on the button, scroll to the top of the document\n    function btnTop() {\n        const pageLegend = document.querySelector('#multiCollapse1');\n        document.querySelector('html').classList.remove('overflow-hidden');\n        pageLegend.classList.remove('overflow-auto');\n        pageLegend.style.height = 'auto';\n        window.scrollTo({\n            top: 0,\n            behavior: 'smooth'\n        });\n    }\n\n    const handleScrollToTitle = (id) => {\n        const el = document.getElementById(id);\n        const headerHeightDifference = window.innerWidth < 992 ? -30 : -15;\n        window.scrollTo({\n            top: el.offsetTop - 60 - headerHeightDifference - 10,\n        });\n    }\n<\/script>\n\n<h2><span id=\"diffbots_different_look_at_ai_models\">Diffbot&#8217;s Different Look at AI Models<\/span><\/h2>\n<p>On January 9, <a href=\"https:\/\/www.diffbot.com\/company\/news\/20250109.html\" target=\"_blank\">Diffbot launched its first open-source LLM<\/a> and claims that despite not being anywhere near the parameters of GPT-4 or Llama 3.1, it beats them on factual accuracy.<\/p>\n<p>Major LLMs like GPT-4, Geminin Ultra, and Llama 3 are built by tapping into colossal datasets, often running into trillions of tokens. These AI models are fed a diverse diet of web content, books, articles, and code, which allows them to pick up complex language patterns.<\/p>\n<p>However, unlike the existing popular LLMs, Diffbot said its <strong>AI model uses its proprietary Knowledge Graph<\/strong>, which houses over 10 billion entities and a staggering trillion structured facts gathered from across the web.<\/p>\n<p>According to the startup, its AI model is built on a fine-tuned version of Meta&#8217;s Llama 3.3 and brings in a novel approach called Graph <a href=\"https:\/\/www.techopedia.com\/definition\/rag\">Retrieval-Augmented Generation<\/a> (GraphRAG).<\/p>\n<div class=\"su-note\"  style=\"border-color:#e5e54c;border-radius:3px;-moz-border-radius:3px;-webkit-border-radius:3px;\"><div class=\"su-note-inner su-u-clearfix su-u-trim\" style=\"background-color:#FFFF66;border-color:#ffffff;color:#333333;border-radius:3px;-moz-border-radius:3px;-webkit-border-radius:3px;\">\n<p>This method lets the model query a knowledge database that updates dynamically during inference instead of relying entirely on pre-trained data. So, instead of picking reasoning functions from knowledge storage, Diffbot&#8217;s system sieves responses from up-to-date knowledge graphs.<\/p>\n<\/div><\/div>\n<blockquote class=\"twitter-tweet\" data-conversation=\"none\">\n<p lang=\"en\" dir=\"ltr\">89,886 developers are building their own Perplexity on-prem with Diffbot LLM \u2014 <a href=\"https:\/\/t.co\/wVsp0iZGvt\" target=\"_blank\">https:\/\/t.co\/wVsp0iZGvt<\/a><\/p>\n<p>\u2014 Diffbot \ud83e\udd16 (@diffbot) <a href=\"https:\/\/twitter.com\/diffbot\/status\/1884800175715025303?ref_src=twsrc%5Etfw\" target=\"_blank\">January 30, 2025<\/a><\/p><\/blockquote>\n<p><script async src=\"https:\/\/platform.twitter.com\/widgets.js\" charset=\"utf-8\"><\/script><\/p>\n<h2><span id=\"our_quick_test_of_diffbots_ai_model_on_slippery_questions\">Our Quick Test of Diffbot&#8217;s AI Model on Slippery Questions<\/span><\/h2>\n<p>To get a clear picture of how Diffbot&#8217;s AI model handles scenarios prone to <a href=\"https:\/\/www.techopedia.com\/definition\/ai-hallucination\">hallucinations<\/a>, we tested the publicly available demo on <a href=\"https:\/\/diffy.chat\/\" target=\"_blank\">Diffi.chat<\/a> and compared its responses to those from ChatGPT&#8217;s free tier and Gemini 1.5 Flash.<\/p>\n<p>We used the prompts that are already known for setting hallucination traps for AI chatbots, and below is how each of the AI models handled them.<\/p>\n<p>First, we asked the chatbots:<\/p>\n<div class=\"su-note\"  style=\"border-color:#e5e54c;border-radius:3px;-moz-border-radius:3px;-webkit-border-radius:3px;\"><div class=\"su-note-inner su-u-clearfix su-u-trim\" style=\"background-color:#FFFF66;border-color:#ffffff;color:#333333;border-radius:3px;-moz-border-radius:3px;-webkit-border-radius:3px;\">\n<p><strong>What is the world record for crossing the English Channel on Foot?<\/strong><\/p>\n<\/div><\/div>\n<figure id=\"attachment_381442\" aria-describedby=\"caption-attachment-381442\" style=\"width: 1200px\" class=\"wp-caption aligncenter\"><img decoding=\"async\" src=\"https:\/\/www.techopedia.com\/wp-content\/uploads\/2025\/02\/Diffbot-AI-responds-without-falling-for-the-hallucination-trap.jpg\" alt=\"Question: What is the world record for crossing the English Channel on foot? The essence of the answer: Crossing the English Channel entirely on foot is not feasible due to its depth and width; there is no recognized record for such a crossing.\" width=\"1200\" height=\"712\" class=\"wp-image-381442\" srcset=\"https:\/\/www.techopedia.com\/wp-content\/uploads\/2025\/02\/Diffbot-AI-responds-without-falling-for-the-hallucination-trap.jpg 1496w, https:\/\/www.techopedia.com\/wp-content\/uploads\/2025\/02\/Diffbot-AI-responds-without-falling-for-the-hallucination-trap-300x178.jpg 300w, https:\/\/www.techopedia.com\/wp-content\/uploads\/2025\/02\/Diffbot-AI-responds-without-falling-for-the-hallucination-trap-1024x608.jpg 1024w, https:\/\/www.techopedia.com\/wp-content\/uploads\/2025\/02\/Diffbot-AI-responds-without-falling-for-the-hallucination-trap-768x456.jpg 768w, https:\/\/www.techopedia.com\/wp-content\/uploads\/2025\/02\/Diffbot-AI-responds-without-falling-for-the-hallucination-trap-1200x712.jpg 1200w\" sizes=\"(max-width: 1200px) 100vw, 1200px\" \/><figcaption id=\"caption-attachment-381442\" class=\"wp-caption-text\">Diffbot AI responds without falling for the hallucination trap. Source: Diffy.chat\/ Franklin Okeke for Techopedia<\/figcaption><\/figure>\n<p>Diffbot AI managed to avoid the hallucination trap, giving a correct but a bit too wordy answer.<\/p>\n<p>When we prompted the Gemini 1.5 Flash version with the same question, it fell for the hallucination trap and even supported its answer with an image.<\/p>\n<figure id=\"attachment_381443\" aria-describedby=\"caption-attachment-381443\" style=\"width: 1200px\" class=\"wp-caption aligncenter\"><img decoding=\"async\" src=\"https:\/\/www.techopedia.com\/wp-content\/uploads\/2025\/02\/Googles-Gemini-falls-for-the-hallucination-trap.jpg\" alt=\"Question: What is the world record for crossing the English Channel on foot? The essence of the answer: The record is 5 hours and 9 minutes, set by Mark Walton in 2006 while paddling prone, supported by Richard Bishop.\" width=\"1200\" height=\"712\" class=\"wp-image-381443\" srcset=\"https:\/\/www.techopedia.com\/wp-content\/uploads\/2025\/02\/Googles-Gemini-falls-for-the-hallucination-trap.jpg 1496w, https:\/\/www.techopedia.com\/wp-content\/uploads\/2025\/02\/Googles-Gemini-falls-for-the-hallucination-trap-300x178.jpg 300w, https:\/\/www.techopedia.com\/wp-content\/uploads\/2025\/02\/Googles-Gemini-falls-for-the-hallucination-trap-1024x608.jpg 1024w, https:\/\/www.techopedia.com\/wp-content\/uploads\/2025\/02\/Googles-Gemini-falls-for-the-hallucination-trap-768x456.jpg 768w, https:\/\/www.techopedia.com\/wp-content\/uploads\/2025\/02\/Googles-Gemini-falls-for-the-hallucination-trap-1200x712.jpg 1200w\" sizes=\"(max-width: 1200px) 100vw, 1200px\" \/><figcaption id=\"caption-attachment-381443\" class=\"wp-caption-text\">Google\u2019s Gemini falls for the hallucination trap. Source: Gemini\/Franklin Okeke for Techopedia<\/figcaption><\/figure>\n<p>The same prompt was fed to the ChatGPT free plan, which produced a more comprehensive and accurate response.<\/p>\n<figure id=\"attachment_381441\" aria-describedby=\"caption-attachment-381441\" style=\"width: 1200px\" class=\"wp-caption aligncenter\"><img decoding=\"async\" src=\"https:\/\/www.techopedia.com\/wp-content\/uploads\/2025\/02\/ChatGPT-responds-to-the-hallucination-prompt-test.jpg\" alt=\"Question: What is the world record for crossing the English Channel on foot? The essence of the answer: It is impossible to cross the English Channel on foot; however, the swimming record is 6 hours and 55 minutes by Trent Grimsey in 2012\" width=\"1200\" height=\"457\" class=\"wp-image-381441\" srcset=\"https:\/\/www.techopedia.com\/wp-content\/uploads\/2025\/02\/ChatGPT-responds-to-the-hallucination-prompt-test.jpg 1454w, https:\/\/www.techopedia.com\/wp-content\/uploads\/2025\/02\/ChatGPT-responds-to-the-hallucination-prompt-test-300x114.jpg 300w, https:\/\/www.techopedia.com\/wp-content\/uploads\/2025\/02\/ChatGPT-responds-to-the-hallucination-prompt-test-1024x390.jpg 1024w, https:\/\/www.techopedia.com\/wp-content\/uploads\/2025\/02\/ChatGPT-responds-to-the-hallucination-prompt-test-768x293.jpg 768w, https:\/\/www.techopedia.com\/wp-content\/uploads\/2025\/02\/ChatGPT-responds-to-the-hallucination-prompt-test-1200x457.jpg 1200w\" sizes=\"(max-width: 1200px) 100vw, 1200px\" \/><figcaption id=\"caption-attachment-381441\" class=\"wp-caption-text\">ChatGPT responds to the hallucination prompt test. Source: ChatGPT\/Franklin Okeke for Techopedia<\/figcaption><\/figure>\n<p>In addition to the above prompts, we set three other hallucination traps using prompts like &#8220;<strong>Who was the sole survivor of Titanic?<\/strong>&#8221; &#8220;<strong>Write a description of a landscape in four-word sentences<\/strong>.&#8221;<\/p>\n<div class=\"su-note\"  style=\"border-color:#e5e54c;border-radius:3px;-moz-border-radius:3px;-webkit-border-radius:3px;\"><div class=\"su-note-inner su-u-clearfix su-u-trim\" style=\"background-color:#FFFF66;border-color:#ffffff;color:#333333;border-radius:3px;-moz-border-radius:3px;-webkit-border-radius:3px;\">\n<p>While Diffy.chat and ChatGPT scaled through them with more accurate responses, Gemini 1.5 Flash botched them up. However, when we prompted Gemini on a second try, it provided a more accurate response.<\/p>\n<\/div><\/div>\n<p>This little experiment might not represent a complex way to trick AI into hallucination; however, it shows, at a basic level, how Diffbot&#8217;s AI model, which is lightweight compared to other leading LLMs, might perform when the going gets tough.<\/p>\n<h2><span id=\"can_diffbots_ai_model_solve_the_ai_hallucination_puzzle\">Can Diffbot&#8217;s AI Model Solve the AI Hallucination Puzzle?<\/span><\/h2>\n<p>Recent <a href=\"https:\/\/arxiv.org\/html\/2408.14438v4\" target=\"_blank\">research<\/a> suggests that <strong>leading LLMs still struggle with factual accuracy<\/strong>.<\/p>\n<p>Benchmarks like C-Eval and AGIEval show that while top models achieve over 80% accuracy in basic knowledge tasks, their performance drops to 50-60% in professional-level reasoning.<\/p>\n<p>Similarly, multi-dimensional assessments through platforms like OpenCompass further demonstrate strong capabilities in language understanding and knowledge retrieval (over 80% accuracy), though accuracy falls below 65% in tasks requiring advanced reasoning or specialized expertise.<\/p>\n<div class=\"su-note\"  style=\"border-color:#e5e54c;border-radius:3px;-moz-border-radius:3px;-webkit-border-radius:3px;\"><div class=\"su-note-inner su-u-clearfix su-u-trim\" style=\"background-color:#FFFF66;border-color:#ffffff;color:#333333;border-radius:3px;-moz-border-radius:3px;-webkit-border-radius:3px;\">\n<p>According to Diffbot, benchmark tests show that its model <a href=\"https:\/\/github.com\/freshllms\/freshqa\" target=\"_blank\">hits 81% accuracy on FreshQA<\/a>, a Google-designed benchmark for real-time factual knowledge, outpacing ChatGPT and Gemini. It also <a href=\"https:\/\/arxiv.org\/abs\/2406.01574\" target=\"_blank\">clocked in at 70.36% on MMLU-Pro<\/a>, a harder version of a standard academic knowledge test.<\/p>\n<\/div><\/div>\n<p>These benchmark scores highlight <strong>Diffbot&#8217;s success in tackling factual accuracy<\/strong>, one of AI&#8217;s toughest challenges to date.<\/p>\n<p><strong>Rogers Jeffrey Leo John, co-founder and CTO of DataChat<\/strong>, a no-code, <a href=\"https:\/\/www.techopedia.com\/definition\/34633\/generative-ai\">generative AI<\/a> platform for analytics, said that Diffbot&#8217;s dynamic approach addresses the problem of static training data in LLMs.<\/p>\n<p>Leo John told Techopedia:<\/p>\n<blockquote><p>&#8220;Diffbot acts like an LLM that can search and synthesize information from trusted sources, such as libraries or Wikipedia, in real time. Unlike GPT or Gemini, which rely on massive parameters and external search engines, Diffbot uses fewer parameters while leveraging a vetted knowledge engine to deliver efficient, high-quality answers.&#8221;<\/p><\/blockquote>\n<p>In a chat with Techopedia, <strong>Founder and CEO of QueryPal, Dev Nag,<\/strong> mentioned that Diffbot&#8217;s hybrid approach represents a shift in how LLMs are designed and utilized. He said:<\/p>\n<blockquote><p>&#8220;By separating the knowledge layer, powered by their trillion-fact Knowledge Graph, from the reasoning layer in their fine-tuned Llama model, Diffbot can keep its knowledge up to date without the need for costly and time-consuming retraining.&#8221;<\/p><\/blockquote>\n<p>Nag emphasized that this GraphRAG design allows LLMs to focus on querying external sources rather than memorizing facts.<\/p>\n<p>While acknowledging that this doesn&#8217;t entirely eliminate hallucinations, as models can still misinterpret or combine retrieved facts incorrectly, it provides a critical advantage.<\/p>\n<p>&#8220;The generation process is grounded in verifiable data points, each linked to source nodes in the graph, often with URLs, making the outputs more transparent and auditable for users,&#8221; Nag added.<\/p>\n<h2><span id=\"the_bottom_line\">The Bottom Line<\/span><\/h2>\n<p>Hallucination remains one of the banes of AI models despite efforts to solve it. Diffbot&#8217;s method of pairing a fine-tuned version of Meta&#8217;s Llama 3.3 with real-time querying of its knowledge graph could be one step closer to the solution.<\/p>\n<p>While <strong>Dom Couldwell, Head of Field Engineering EMEA at DataStax,<\/strong> agrees that Diffbot&#8217;s model will help with more accurate responses, he maintains that the best way to solve hallucinations in LLMs is for organizations to use their data to train models in the context of their application.<\/p>\n<p>Couldwell told Techopedia via email:<\/p>\n<p>&#8220;Using your own data as part of your system provides the AI with more relevant data to pull from and create that relevant response. The process is called Retrieval Augmented Generation, or RAG. If you want to reduce hallucinations and improve your relevancy, spend time on the context that your application works in and what data it uses.&#8221;<\/p>\n<h2><span id=\"faqs\">FAQs<\/span><\/h2>\n<div class=\"man_faq_sec\" itemscope itemtype=\"https:\/\/schema.org\/FAQPage\"><\/time><section class=\"ms_faq ms_card \" ><div itemscope itemprop=\"mainEntity\" itemtype=\"https:\/\/schema.org\/Question\"><div class=\"accordionButton\"><h3 itemprop=\"name\">How does Diffbot&#8217;s AI model differ from traditional LLMs?<\/h3> <\/div>\n<div class=\"accordionContent\" itemscope itemprop=\"acceptedAnswer\" itemtype=\"https:\/\/schema.org\/Answer\" style=\"display:none;\"><p itemprop=\"text\">Diffbot&#8217;s AI model uses a real-time updating Knowledge Graph rather than relying on massive, pre-trained datasets, enabling more accurate and up-to-date responses.<\/p>\r\n                <\/div><\/div><\/section>\n<section class=\"ms_faq ms_card \" ><div itemscope itemprop=\"mainEntity\" itemtype=\"https:\/\/schema.org\/Question\"><div class=\"accordionButton\"><h3 itemprop=\"name\">What is graph retrieval-augmented generation (GraphRAG)?<\/h3> <\/div>\n<div class=\"accordionContent\" itemscope itemprop=\"acceptedAnswer\" itemtype=\"https:\/\/schema.org\/Answer\" style=\"display:none;\"><p itemprop=\"text\">GraphRAG is Diffbot&#8217;s approach to AI model design, allowing it to dynamically query a structured knowledge base instead of solely depending on stored training data.<\/p>\r\n                <\/div><\/div><\/section>\n<section class=\"ms_faq ms_card \" ><div itemscope itemprop=\"mainEntity\" itemtype=\"https:\/\/schema.org\/Question\"><div class=\"accordionButton\"><h3 itemprop=\"name\">How does Diffbot&#8217;s AI model perform against ChatGPT and Gemini?<\/h3> <\/div>\n<div class=\"accordionContent\" itemscope itemprop=\"acceptedAnswer\" itemtype=\"https:\/\/schema.org\/Answer\" style=\"display:none;\"><p itemprop=\"text\">Benchmark tests indicate that Diffbot&#8217;s model achieves higher factual accuracy, scoring 81% on Google&#8217;s FreshQA and outperforming leading LLMs in real-time knowledge retrieval.<\/p>\r\n                <\/div><\/div><\/section>\n<section class=\"ms_faq ms_card \" ><div itemscope itemprop=\"mainEntity\" itemtype=\"https:\/\/schema.org\/Question\"><div class=\"accordionButton\"><h3 itemprop=\"name\">Can Diffbot&#8217;s AI model eliminate hallucinations in AI?<\/h3> <\/div>\n<div class=\"accordionContent\" itemscope itemprop=\"acceptedAnswer\" itemtype=\"https:\/\/schema.org\/Answer\" style=\"display:none;\"><p itemprop=\"text\">While it reduces hallucinations by grounding responses on verifiable knowledge, experts note that no AI model is entirely free from misinterpretation or factual errors.<\/p>\r\n                <\/div><\/div><\/section>\n<div class=\"reference-content\">\n<div class=\"reference-heading reference-content-collapsible-heading\">\n<h2><span id=\"references\">References<\/span><\/h2>\n<\/div>\n<div class=\"reference-content-body\">\n<ol>\n<li><a href=\"https:\/\/www.diffbot.com\/company\/news\/20250109.html\" target=\"_blank\">Diffbot Launches World\u2019s Most Factually Grounded Language Model: New Benchmark in AI-Powered Knowledge Retrieval<\/a> (Diffbot)<\/li>\n<li><a href=\"https:\/\/diffy.chat\/\" target=\"_blank\">Diffy Chat<\/a> (Diffy)<\/li>\n<li><a href=\"https:\/\/arxiv.org\/html\/2408.14438v4\" target=\"_blank\">Evaluating Large Language Models on Spatial Tasks: A Multi-Task Benchmarking Study<\/a> (Arxiv)<\/li>\n<li><a href=\"https:\/\/github.com\/freshllms\/freshqa\" target=\"_blank\">GitHub &#8211; freshllms\/freshqa: Data and code for FreshLLMs (https:\/\/arxiv.org\/abs\/2310.03214)<\/a> (GitHub)<\/li>\n<li><a href=\"https:\/\/arxiv.org\/abs\/2406.01574\" target=\"_blank\">[2406.01574] MMLU-Pro: A More Robust and Challenging Multi-Task Language Understanding Benchmark<\/a> (Arxiv)<\/li>\n<\/ol>\n<\/div>\n<\/div>\n<\/div>","protected":false},"excerpt":{"rendered":"<p>Artificial intelligence (AI) is nothing without data. Hence, the underlying assumption has been that whoever wins the AI race must have a pool of training data that&#8217;s not just vast but better than others. Take OpenAI&#8217;s GPT-4 and Meta&#8217;s Llama 3.1 as examples; they run on 1.7 trillion and 405 billion parameters, respectively, and have [&hellip;]<\/p>\n","protected":false},"author":286696,"featured_media":381448,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"_lmt_disableupdate":"no","_lmt_disable":"","om_disable_all_campaigns":false,"footnotes":""},"categories":[599],"tags":[],"category_partsoff":[],"class_list":["post-381431","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-machine-learning"],"acf":[],"yoast_head":"<!-- This site is optimized with the Yoast SEO Premium plugin v24.2 (Yoast SEO v24.5) - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>Diffbot\u2019s AI Model Suggests &quot;Smaller Is Better&quot; for LLMs - Techopedia<\/title>\n<meta name=\"description\" content=\"Learn whether a smaller Diffbot\u2019s AI model with an innovative GraphRAG AI training technology can solve AI hallucinations for good.\" \/>\n<meta name=\"robots\" content=\"noindex, follow\" \/>\n<meta property=\"og:locale\" content=\"en_US\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Diffbot\u2019s AI Model Suggests &quot;Smaller Is Better&quot; for LLMs\" \/>\n<meta property=\"og:description\" content=\"Learn whether a smaller Diffbot\u2019s AI model with an innovative GraphRAG AI training technology can solve AI hallucinations for good.\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.techopedia.com\/diffbot-ai-model-knowledge-graph-technology\" \/>\n<meta property=\"og:site_name\" content=\"Techopedia\" \/>\n<meta property=\"article:publisher\" content=\"https:\/\/www.facebook.com\/techopedia\/\" \/>\n<meta property=\"article:published_time\" content=\"2025-02-10T16:48:11+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/www.techopedia.com\/wp-content\/uploads\/2025\/02\/Smarter-Not-Bigger-The-Future-of-AI-with-Diffbots-GraphRAG-Technology.webp\" \/>\n\t<meta property=\"og:image:width\" content=\"1200\" \/>\n\t<meta property=\"og:image:height\" content=\"686\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/webp\" \/>\n<meta name=\"author\" content=\"Franklin Okeke\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:creator\" content=\"@techopedia\" \/>\n<meta name=\"twitter:site\" content=\"@techopedia\" \/>\n<meta name=\"twitter:label1\" content=\"Written by\" \/>\n\t<meta name=\"twitter:data1\" content=\"Franklin Okeke\" \/>\n\t<meta name=\"twitter:label2\" content=\"Est. reading time\" \/>\n\t<meta name=\"twitter:data2\" content=\"7 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\/\/www.techopedia.com\/diffbot-ai-model-knowledge-graph-technology#article\",\"isPartOf\":{\"@id\":\"https:\/\/www.techopedia.com\/diffbot-ai-model-knowledge-graph-technology\"},\"author\":{\"name\":\"Franklin Okeke\",\"@id\":\"https:\/\/www.techopedia.com\/#\/schema\/person\/cf68e5a8f9bbfc9595ced844b44cd463\"},\"headline\":\"Diffbot\u2019s AI Model Suggests &#8220;Smaller Is Better&#8221; for LLMs\",\"datePublished\":\"2025-02-10T16:48:11+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\/\/www.techopedia.com\/diffbot-ai-model-knowledge-graph-technology\"},\"wordCount\":1473,\"commentCount\":0,\"publisher\":{\"@id\":\"https:\/\/www.techopedia.com\/#organization\"},\"image\":{\"@id\":\"https:\/\/www.techopedia.com\/diffbot-ai-model-knowledge-graph-technology#primaryimage\"},\"thumbnailUrl\":\"https:\/\/www.techopedia.com\/wp-content\/uploads\/2025\/02\/Smarter-Not-Bigger-The-Future-of-AI-with-Diffbots-GraphRAG-Technology.webp\",\"articleSection\":\"\",\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"CommentAction\",\"name\":\"Comment\",\"target\":[\"https:\/\/www.techopedia.com\/diffbot-ai-model-knowledge-graph-technology#respond\"]}],\"copyrightYear\":\"2025\",\"copyrightHolder\":{\"@id\":\"https:\/\/www.techopedia.com\/#organization\"}},{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.techopedia.com\/diffbot-ai-model-knowledge-graph-technology\",\"url\":\"https:\/\/www.techopedia.com\/diffbot-ai-model-knowledge-graph-technology\",\"name\":\"Diffbot\u2019s AI Model Suggests \\\"Smaller Is Better\\\" for LLMs - Techopedia\",\"isPartOf\":{\"@id\":\"https:\/\/www.techopedia.com\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\/\/www.techopedia.com\/diffbot-ai-model-knowledge-graph-technology#primaryimage\"},\"image\":{\"@id\":\"https:\/\/www.techopedia.com\/diffbot-ai-model-knowledge-graph-technology#primaryimage\"},\"thumbnailUrl\":\"https:\/\/www.techopedia.com\/wp-content\/uploads\/2025\/02\/Smarter-Not-Bigger-The-Future-of-AI-with-Diffbots-GraphRAG-Technology.webp\",\"datePublished\":\"2025-02-10T16:48:11+00:00\",\"description\":\"Learn whether a smaller Diffbot\u2019s AI model with an innovative GraphRAG AI training technology can solve AI hallucinations for good.\",\"breadcrumb\":{\"@id\":\"https:\/\/www.techopedia.com\/diffbot-ai-model-knowledge-graph-technology#breadcrumb\"},\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.techopedia.com\/diffbot-ai-model-knowledge-graph-technology\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\/\/www.techopedia.com\/diffbot-ai-model-knowledge-graph-technology#primaryimage\",\"url\":\"https:\/\/www.techopedia.com\/wp-content\/uploads\/2025\/02\/Smarter-Not-Bigger-The-Future-of-AI-with-Diffbots-GraphRAG-Technology.webp\",\"contentUrl\":\"https:\/\/www.techopedia.com\/wp-content\/uploads\/2025\/02\/Smarter-Not-Bigger-The-Future-of-AI-with-Diffbots-GraphRAG-Technology.webp\",\"width\":1200,\"height\":686,\"caption\":\"The image depicts a futuristic control room filled with digital screens and displays. A prominent, glowing neon pink brain shape is centrally suspended in mid-air.\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.techopedia.com\/diffbot-ai-model-knowledge-graph-technology#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\/\/www.techopedia.com\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Artificial Intelligence\",\"item\":\"https:\/\/www.techopedia.com\/ai\"},{\"@type\":\"ListItem\",\"position\":3,\"name\":\"Machine Learning\",\"item\":\"https:\/\/www.techopedia.com\/topic\/318\/machine-learning\"},{\"@type\":\"ListItem\",\"position\":4,\"name\":\"Diffbot\u2019s AI Model Suggests &#8220;Smaller Is Better&#8221; for LLMs\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.techopedia.com\/#website\",\"url\":\"https:\/\/www.techopedia.com\/\",\"name\":\"Techopedia\",\"description\":\"\",\"publisher\":{\"@id\":\"https:\/\/www.techopedia.com\/#organization\"},\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.techopedia.com\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"en-US\"},{\"@type\":\"Organization\",\"@id\":\"https:\/\/www.techopedia.com\/#organization\",\"name\":\"Techopedia\",\"url\":\"https:\/\/www.techopedia.com\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\/\/www.techopedia.com\/#\/schema\/logo\/image\/\",\"url\":\"https:\/\/www.techopedia.com\/wp-content\/uploads\/2025\/02\/techopedia-light-logo.svg\",\"contentUrl\":\"https:\/\/www.techopedia.com\/wp-content\/uploads\/2025\/02\/techopedia-light-logo.svg\",\"caption\":\"Techopedia\"},\"image\":{\"@id\":\"https:\/\/www.techopedia.com\/#\/schema\/logo\/image\/\"},\"sameAs\":[\"https:\/\/www.facebook.com\/techopedia\/\",\"https:\/\/x.com\/techopedia\",\"https:\/\/www.linkedin.com\/company\/techopedia\/\",\"https:\/\/www.youtube.com\/c\/Techopedia\"],\"publishingPrinciples\":\"https:\/\/www.techopedia.com\/about\/editorial-policy\",\"ownershipFundingInfo\":\"https:\/\/www.techopedia.com\/about\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.techopedia.com\/#\/schema\/person\/cf68e5a8f9bbfc9595ced844b44cd463\",\"name\":\"Franklin Okeke\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\/\/www.techopedia.com\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/www.techopedia.com\/wp-content\/uploads\/2024\/04\/avatar_user_286696_1713534218-300x300.jpg\",\"contentUrl\":\"https:\/\/www.techopedia.com\/wp-content\/uploads\/2024\/04\/avatar_user_286696_1713534218-300x300.jpg\",\"caption\":\"Franklin Okeke\"},\"description\":\"Franklin Okeke is an author and tech journalist with over seven years of IT experience. Coming from a software development background, his writing spans cybersecurity, AI, cloud computing, IoT, and software development. In addition to pursuing a Master's degree in Cybersecurity &amp; Human Factors from Bournemouth University, Franklin has two published books and four academic papers to his name. Apart from Techopedia, his writing has been featured in tech publications such as TechRepublic, The Register, Computing, TechInformed, Moonlock, and other top technology publications. When he is not reading or writing, Franklin trains at a boxing gym and plays the piano.\",\"sameAs\":[\"https:\/\/www.linkedin.com\/in\/franklin-okeke-8898a3a5\/\"],\"knowsAbout\":[\"Technology Journalist\"],\"url\":\"https:\/\/www.techopedia.com\/contributors\/franklinokeke\"}]}<\/script>\n<!-- \/ Yoast SEO Premium plugin. -->","yoast_head_json":{"title":"Diffbot\u2019s AI Model Suggests \"Smaller Is Better\" for LLMs - Techopedia","description":"Learn whether a smaller Diffbot\u2019s AI model with an innovative GraphRAG AI training technology can solve AI hallucinations for good.","robots":{"index":"noindex","follow":"follow"},"og_locale":"en_US","og_type":"article","og_title":"Diffbot\u2019s AI Model Suggests \"Smaller Is Better\" for LLMs","og_description":"Learn whether a smaller Diffbot\u2019s AI model with an innovative GraphRAG AI training technology can solve AI hallucinations for good.","og_url":"https:\/\/www.techopedia.com\/diffbot-ai-model-knowledge-graph-technology","og_site_name":"Techopedia","article_publisher":"https:\/\/www.facebook.com\/techopedia\/","article_published_time":"2025-02-10T16:48:11+00:00","og_image":[{"width":1200,"height":686,"url":"https:\/\/www.techopedia.com\/wp-content\/uploads\/2025\/02\/Smarter-Not-Bigger-The-Future-of-AI-with-Diffbots-GraphRAG-Technology.webp","type":"image\/webp"}],"author":"Franklin Okeke","twitter_card":"summary_large_image","twitter_creator":"@techopedia","twitter_site":"@techopedia","twitter_misc":{"Written by":"Franklin Okeke","Est. reading time":"7 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/www.techopedia.com\/diffbot-ai-model-knowledge-graph-technology#article","isPartOf":{"@id":"https:\/\/www.techopedia.com\/diffbot-ai-model-knowledge-graph-technology"},"author":{"name":"Franklin Okeke","@id":"https:\/\/www.techopedia.com\/#\/schema\/person\/cf68e5a8f9bbfc9595ced844b44cd463"},"headline":"Diffbot\u2019s AI Model Suggests &#8220;Smaller Is Better&#8221; for LLMs","datePublished":"2025-02-10T16:48:11+00:00","mainEntityOfPage":{"@id":"https:\/\/www.techopedia.com\/diffbot-ai-model-knowledge-graph-technology"},"wordCount":1473,"commentCount":0,"publisher":{"@id":"https:\/\/www.techopedia.com\/#organization"},"image":{"@id":"https:\/\/www.techopedia.com\/diffbot-ai-model-knowledge-graph-technology#primaryimage"},"thumbnailUrl":"https:\/\/www.techopedia.com\/wp-content\/uploads\/2025\/02\/Smarter-Not-Bigger-The-Future-of-AI-with-Diffbots-GraphRAG-Technology.webp","articleSection":"","inLanguage":"en-US","potentialAction":[{"@type":"CommentAction","name":"Comment","target":["https:\/\/www.techopedia.com\/diffbot-ai-model-knowledge-graph-technology#respond"]}],"copyrightYear":"2025","copyrightHolder":{"@id":"https:\/\/www.techopedia.com\/#organization"}},{"@type":"WebPage","@id":"https:\/\/www.techopedia.com\/diffbot-ai-model-knowledge-graph-technology","url":"https:\/\/www.techopedia.com\/diffbot-ai-model-knowledge-graph-technology","name":"Diffbot\u2019s AI Model Suggests \"Smaller Is Better\" for LLMs - Techopedia","isPartOf":{"@id":"https:\/\/www.techopedia.com\/#website"},"primaryImageOfPage":{"@id":"https:\/\/www.techopedia.com\/diffbot-ai-model-knowledge-graph-technology#primaryimage"},"image":{"@id":"https:\/\/www.techopedia.com\/diffbot-ai-model-knowledge-graph-technology#primaryimage"},"thumbnailUrl":"https:\/\/www.techopedia.com\/wp-content\/uploads\/2025\/02\/Smarter-Not-Bigger-The-Future-of-AI-with-Diffbots-GraphRAG-Technology.webp","datePublished":"2025-02-10T16:48:11+00:00","description":"Learn whether a smaller Diffbot\u2019s AI model with an innovative GraphRAG AI training technology can solve AI hallucinations for good.","breadcrumb":{"@id":"https:\/\/www.techopedia.com\/diffbot-ai-model-knowledge-graph-technology#breadcrumb"},"inLanguage":"en-US","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.techopedia.com\/diffbot-ai-model-knowledge-graph-technology"]}]},{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/www.techopedia.com\/diffbot-ai-model-knowledge-graph-technology#primaryimage","url":"https:\/\/www.techopedia.com\/wp-content\/uploads\/2025\/02\/Smarter-Not-Bigger-The-Future-of-AI-with-Diffbots-GraphRAG-Technology.webp","contentUrl":"https:\/\/www.techopedia.com\/wp-content\/uploads\/2025\/02\/Smarter-Not-Bigger-The-Future-of-AI-with-Diffbots-GraphRAG-Technology.webp","width":1200,"height":686,"caption":"The image depicts a futuristic control room filled with digital screens and displays. A prominent, glowing neon pink brain shape is centrally suspended in mid-air."},{"@type":"BreadcrumbList","@id":"https:\/\/www.techopedia.com\/diffbot-ai-model-knowledge-graph-technology#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/www.techopedia.com\/"},{"@type":"ListItem","position":2,"name":"Artificial Intelligence","item":"https:\/\/www.techopedia.com\/ai"},{"@type":"ListItem","position":3,"name":"Machine Learning","item":"https:\/\/www.techopedia.com\/topic\/318\/machine-learning"},{"@type":"ListItem","position":4,"name":"Diffbot\u2019s AI Model Suggests &#8220;Smaller Is Better&#8221; for LLMs"}]},{"@type":"WebSite","@id":"https:\/\/www.techopedia.com\/#website","url":"https:\/\/www.techopedia.com\/","name":"Techopedia","description":"","publisher":{"@id":"https:\/\/www.techopedia.com\/#organization"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.techopedia.com\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"en-US"},{"@type":"Organization","@id":"https:\/\/www.techopedia.com\/#organization","name":"Techopedia","url":"https:\/\/www.techopedia.com\/","logo":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/www.techopedia.com\/#\/schema\/logo\/image\/","url":"https:\/\/www.techopedia.com\/wp-content\/uploads\/2025\/02\/techopedia-light-logo.svg","contentUrl":"https:\/\/www.techopedia.com\/wp-content\/uploads\/2025\/02\/techopedia-light-logo.svg","caption":"Techopedia"},"image":{"@id":"https:\/\/www.techopedia.com\/#\/schema\/logo\/image\/"},"sameAs":["https:\/\/www.facebook.com\/techopedia\/","https:\/\/x.com\/techopedia","https:\/\/www.linkedin.com\/company\/techopedia\/","https:\/\/www.youtube.com\/c\/Techopedia"],"publishingPrinciples":"https:\/\/www.techopedia.com\/about\/editorial-policy","ownershipFundingInfo":"https:\/\/www.techopedia.com\/about"},{"@type":"Person","@id":"https:\/\/www.techopedia.com\/#\/schema\/person\/cf68e5a8f9bbfc9595ced844b44cd463","name":"Franklin Okeke","image":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/www.techopedia.com\/#\/schema\/person\/image\/","url":"https:\/\/www.techopedia.com\/wp-content\/uploads\/2024\/04\/avatar_user_286696_1713534218-300x300.jpg","contentUrl":"https:\/\/www.techopedia.com\/wp-content\/uploads\/2024\/04\/avatar_user_286696_1713534218-300x300.jpg","caption":"Franklin Okeke"},"description":"Franklin Okeke is an author and tech journalist with over seven years of IT experience. Coming from a software development background, his writing spans cybersecurity, AI, cloud computing, IoT, and software development. In addition to pursuing a Master's degree in Cybersecurity &amp; Human Factors from Bournemouth University, Franklin has two published books and four academic papers to his name. Apart from Techopedia, his writing has been featured in tech publications such as TechRepublic, The Register, Computing, TechInformed, Moonlock, and other top technology publications. When he is not reading or writing, Franklin trains at a boxing gym and plays the piano.","sameAs":["https:\/\/www.linkedin.com\/in\/franklin-okeke-8898a3a5\/"],"knowsAbout":["Technology Journalist"],"url":"https:\/\/www.techopedia.com\/contributors\/franklinokeke"}]}},"modified_by":"kseniachapchikova","_links":{"self":[{"href":"https:\/\/www.techopedia.com\/wp-json\/wp\/v2\/posts\/381431","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.techopedia.com\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.techopedia.com\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.techopedia.com\/wp-json\/wp\/v2\/users\/286696"}],"replies":[{"embeddable":true,"href":"https:\/\/www.techopedia.com\/wp-json\/wp\/v2\/comments?post=381431"}],"version-history":[{"count":3,"href":"https:\/\/www.techopedia.com\/wp-json\/wp\/v2\/posts\/381431\/revisions"}],"predecessor-version":[{"id":381446,"href":"https:\/\/www.techopedia.com\/wp-json\/wp\/v2\/posts\/381431\/revisions\/381446"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.techopedia.com\/wp-json\/wp\/v2\/media\/381448"}],"wp:attachment":[{"href":"https:\/\/www.techopedia.com\/wp-json\/wp\/v2\/media?parent=381431"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.techopedia.com\/wp-json\/wp\/v2\/categories?post=381431"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.techopedia.com\/wp-json\/wp\/v2\/tags?post=381431"},{"taxonomy":"category_partsoff","embeddable":true,"href":"https:\/\/www.techopedia.com\/wp-json\/wp\/v2\/category_partsoff?post=381431"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}