{"id":1378,"date":"2024-11-11T10:00:00","date_gmt":"2024-11-11T10:00:00","guid":{"rendered":"https:\/\/netguide.io\/news\/2024\/11\/11\/mit-schlaegt-mit-neuem-llm-den-rekord-im-agi-pub-benchmark-um-20\/"},"modified":"2026-08-17T23:21:09","modified_gmt":"2026-08-17T23:21:09","slug":"mit-schlaegt-mit-neuem-llm-den-rekord-im-agi-pub-benchmark-um-20","status":"publish","type":"post","link":"https:\/\/netguide.io\/news\/de\/2024\/11\/11\/mit-schlaegt-mit-neuem-llm-den-rekord-im-agi-pub-benchmark-um-20\/","title":{"rendered":"MIT schl\u00e4gt mit neuem LLM den Rekord im ARC-AGI-PUB Benchmark um rund 20 Prozent"},"content":{"rendered":"<div id=\"netgu-3563660110\" class=\"netgu-before-content netgu-entity-placement\"><script async src=\"\/\/pagead2.googlesyndication.com\/pagead\/js\/adsbygoogle.js?client=ca-pub-6258556257245998\" crossorigin=\"anonymous\"><\/script><ins class=\"adsbygoogle\" style=\"display:block;\" data-ad-client=\"ca-pub-6258556257245998\" \ndata-ad-slot=\"3494115342\" \ndata-ad-format=\"auto\"><\/ins>\n<script> \n(adsbygoogle = window.adsbygoogle || []).push({}); \n<\/script>\n<\/div>\n<p class=\"wp-block-paragraph\">Ein Team vom MIT hat mithilfe eines Sprachmodells (Large Language Model, LLM) mit 8 Milliarden Parametern und einer Technik namens Test-Time Training (TTT) einen bemerkenswerten Fortschritt in der abstrakten Aufgabenl\u00f6sung erzielt. Mit einer Leistung von 61,9 % auf dem ARC-AGI-PUB Benchmark \u00fcbertraf das Team den bisherigen Rekord von 42 % deutlich. Das Ergebnis wiegt schwer, weil ARC-AGI-PUB eine anspruchsvolle Sammlung von Aufgaben ist, die visuelle Mustererkennung und komplexes logisches Denken verlangt.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Was ist Test-Time Training (TTT)?<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Test-Time Training erlaubt es dem Modell, w\u00e4hrend der Testphase kleine Anpassungen vorzunehmen, um auf neue Aufgaben zu reagieren. Dabei nutzt TTT den Kontext der Eingabedaten, um das Modell tempor\u00e4r f\u00fcr die jeweils aktuelle Aufgabe zu optimieren. Das unterscheidet sich von klassischen Ans\u00e4tzen, bei denen ein Modell ausschlie\u00dflich vorab trainiert wird. Gerade bei komplexen und neuartigen Aufgaben zeigt TTT deutliche Leistungsverbesserungen.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Wie gelang der Durchbruch am MIT?<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Entscheidend war laut dem Team die Kombination aus einem initialen Feintuning und speziellen Transformationen f\u00fcr die Testdaten. Durch diese Anpassungen verbesserte das Modell seine F\u00e4higkeit zur Mustererkennung. Zus\u00e4tzlich kombinierten die Forschenden das Verfahren in einem Ensemble mit modernen Programmieransatz-Methoden und steigerten den Leistungswert so auf 61,9 % \u2013 ein Niveau, das der Leistung eines durchschnittlichen Menschen entspricht.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Warum die Entwicklung wichtig ist<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Die Ergebnisse zeigen, dass Sprachmodelle abstrakte Aufgaben auch ohne explizite symbolische Methoden l\u00f6sen k\u00f6nnen. Damit r\u00fcckt die Frage in den Fokus, wie weit sich reine LLM-Ans\u00e4tze an menschliche F\u00e4higkeiten in der abstrakten Probleml\u00f6sung ann\u00e4hern k\u00f6nnen. Weitere Details finden sich im <a href=\"https:\/\/ekinakyurek.github.io\/papers\/ttt.pdf\" target=\"_blank\" rel=\"noreferrer noopener\">offiziellen Paper<\/a>.<\/p>\n<div id=\"netgu-1822120542\" class=\"netgu-after-content netgu-entity-placement\"><script async src=\"\/\/pagead2.googlesyndication.com\/pagead\/js\/adsbygoogle.js?client=ca-pub-6258556257245998\" crossorigin=\"anonymous\"><\/script><ins class=\"adsbygoogle\" style=\"display:block;\" data-ad-client=\"ca-pub-6258556257245998\" \ndata-ad-slot=\"4559785002\" \ndata-ad-format=\"auto\"><\/ins>\n<script> \n(adsbygoogle = window.adsbygoogle || []).push({}); \n<\/script>\n<\/div>","protected":false},"excerpt":{"rendered":"<p>Ein MIT-Team erreicht mit einem 8-Milliarden-Parameter-LLM und Test-Time Training 61,9 % im ARC-AGI-PUB Benchmark \u2013 deutlich \u00fcber dem bisherigen Rekord.<\/p>\n","protected":false},"author":1,"featured_media":1624,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"site-sidebar-layout":"default","site-content-layout":"","ast-site-content-layout":"default","site-content-style":"default","site-sidebar-style":"default","ast-global-header-display":"","ast-banner-title-visibility":"","ast-main-header-display":"","ast-hfb-above-header-display":"","ast-hfb-below-header-display":"","ast-hfb-mobile-header-display":"","site-post-title":"","ast-breadcrumbs-content":"","ast-featured-img":"","footer-sml-layout":"","ast-disable-related-posts":"","theme-transparent-header-meta":"","adv-header-id-meta":"","stick-header-meta":"","header-above-stick-meta":"","header-main-stick-meta":"","header-below-stick-meta":"","astra-migrate-meta-layouts":"default","ast-page-background-enabled":"default","ast-page-background-meta":{"desktop":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"ast-content-background-meta":{"desktop":{"background-color":"var(--ast-global-color-4)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"var(--ast-global-color-4)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"var(--ast-global-color-4)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"footnotes":""},"categories":[],"tags":[],"class_list":["post-1378","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry"],"brizy_media":[],"_links":{"self":[{"href":"https:\/\/netguide.io\/news\/wp-json\/wp\/v2\/posts\/1378","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/netguide.io\/news\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/netguide.io\/news\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/netguide.io\/news\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/netguide.io\/news\/wp-json\/wp\/v2\/comments?post=1378"}],"version-history":[{"count":1,"href":"https:\/\/netguide.io\/news\/wp-json\/wp\/v2\/posts\/1378\/revisions"}],"predecessor-version":[{"id":1625,"href":"https:\/\/netguide.io\/news\/wp-json\/wp\/v2\/posts\/1378\/revisions\/1625"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/netguide.io\/news\/wp-json\/wp\/v2\/media\/1624"}],"wp:attachment":[{"href":"https:\/\/netguide.io\/news\/wp-json\/wp\/v2\/media?parent=1378"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/netguide.io\/news\/wp-json\/wp\/v2\/categories?post=1378"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/netguide.io\/news\/wp-json\/wp\/v2\/tags?post=1378"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}