{"id":10342,"date":"2024-02-27T18:58:26","date_gmt":"2024-02-27T18:58:26","guid":{"rendered":"https:\/\/dailyai.com\/?p=10342"},"modified":"2024-03-28T00:37:55","modified_gmt":"2024-03-28T00:37:55","slug":"generative-ai-systems-hallucinations-and-mounting-technical-debt","status":"publish","type":"post","link":"https:\/\/dailyai.com\/da\/2024\/02\/generative-ai-systems-hallucinations-and-mounting-technical-debt\/","title":{"rendered":"Generative AI-systemer, hallucinationer og stigende teknisk g\u00e6ld"},"content":{"rendered":"<p><strong>Efterh\u00e5nden som AI-systemer som store sprogmodeller (LLM'er) vokser i st\u00f8rrelse og kompleksitet, afd\u00e6kker forskere sp\u00e6ndende grundl\u00e6ggende begr\u00e6nsninger.\u00a0<\/strong><\/p>\n<p>Nylige unders\u00f8gelser fra Google og University of Singapore har afd\u00e6kket mekanikken bag AI-\"hallucinationer\" - hvor modeller genererer overbevisende, men fabrikerede oplysninger - og ophobningen af \"teknisk g\u00e6ld\", som kan skabe rodede, up\u00e5lidelige systemer over tid.<\/p>\n<p>Ud over de tekniske udfordringer er det stadig et \u00e5bent sp\u00f8rgsm\u00e5l at afstemme AI's evner og incitamenter med menneskelige v\u00e6rdier.<\/p>\n<p>N\u00e5r virksomheder som OpenAI skubber p\u00e5 mod kunstig generel intelligens (AGI), betyder det at sikre vejen frem at anerkende gr\u00e6nserne for de nuv\u00e6rende systemer.<\/p>\n<p><span style=\"font-weight: 400;\">Men en omhyggelig anerkendelse af risici er i modstrid med Silicon Valleys motto om at \"bev\u00e6ge sig hurtigt og \u00f8del\u00e6gge ting\", som kendetegner F&amp;U inden for kunstig intelligens, ligesom det gjorde for teknologiske innovationer f\u00f8r det.\u00a0<\/span><\/p>\n<h2>Studie 1: AI-modeller oparbejder 'teknisk g\u00e6ld'<\/h2>\n<p><span style=\"font-weight: 400;\">Maskinl\u00e6ring udr\u00e5bes ofte som kontinuerligt skalerbar, og systemerne tilbyder en modul\u00e6r, integreret ramme for udvikling.\u00a0<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Men i baggrunden oparbejder udviklerne m\u00e5ske en h\u00f8j grad af \"teknisk g\u00e6ld\", som de bliver n\u00f8dt til at l\u00f8se hen ad vejen.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">I en <\/span><a href=\"https:\/\/storage.googleapis.com\/gweb-research2023-media\/pubtools\/pdf\/43146.pdf\" target=\"_blank\" rel=\"noopener\"><span style=\"font-weight: 400;\">Google-forskningsrapport<\/span><\/a><span style=\"font-weight: 400;\">I artiklen \"Machine Learning: The High-Interest Credit Card of Technical Debt\", diskuterer forskere begrebet teknisk g\u00e6ld i forbindelse med ML-systemer.\u00a0<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Kaggle CEO og mange\u00e5rig Google-forsker D. Sculley og kolleger h\u00e6vder, at selvom ML tilbyder st\u00e6rke v\u00e6rkt\u00f8jer til hurtigt at opbygge komplekse systemer, er disse \"quick wins\" ofte misvisende.\u00a0<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Den enkle og hurtige implementering af ML-modeller kan skjule de fremtidige byrder, de p\u00e5f\u00f8rer systemets vedligeholdelsesevne og udvikling. <\/span><\/p>\n<p><span style=\"font-weight: 400;\">Som forfatterne beskriver, stammer denne skjulte g\u00e6ld fra flere ML-specifikke risikofaktorer, som udviklere b\u00f8r undg\u00e5 eller refaktorere.<\/span><\/p>\n<p>Her er de vigtigste indsigter:<\/p>\n<ul>\n<li><span style=\"font-weight: 400;\">ML-systemer introducerer i sagens natur et niveau af kompleksitet, der g\u00e5r ud over kodning alene. <\/span>Det kan f\u00f8re til det, forfatterne kalder \"gr\u00e6nseerosion\", hvor de klare linjer mellem forskellige systemkomponenter bliver sl\u00f8rede p\u00e5 grund af den gensidige afh\u00e6ngighed, som ML-modellerne skaber. Det g\u00f8r det vanskeligt at isolere og implementere forbedringer uden at p\u00e5virke andre dele af systemet.<\/li>\n<li>Artiklen fremh\u00e6ver ogs\u00e5 problemet med \"sammenfiltring\", hvor \u00e6ndringer i en hvilken som helst del af et ML-system, f.eks. inputfunktioner eller modelparametre, kan have uforudsigelige effekter p\u00e5 resten af systemet. \u00c6ndring af en lille parameter kan udl\u00f8se en kaskade af effekter, der p\u00e5virker hele modellens funktion og integritet.<\/li>\n<li>Et andet problem er skabelsen af \"skjulte feedbacksl\u00f8jfer\", hvor ML-modeller p\u00e5virker deres egne tr\u00e6ningsdata p\u00e5 uforudsete m\u00e5der. Det kan f\u00f8re til systemer, der udvikler sig i utilsigtede retninger, hvilket g\u00f8r det endnu sv\u00e6rere at styre og forst\u00e5 systemets adf\u00e6rd.<\/li>\n<li>Forfatterne besk\u00e6ftiger sig ogs\u00e5 med \"dataafh\u00e6ngighed\", som f.eks. n\u00e5r inputsignaler \u00e6ndrer sig over tid, hvilket er s\u00e6rligt problematisk, da det er sv\u00e6rere at opdage.<\/li>\n<\/ul>\n<h3>Hvorfor teknisk g\u00e6ld er vigtig<\/h3>\n<p><span style=\"font-weight: 400;\">Teknisk g\u00e6ld ber\u00f8rer ML-systemers sundhed og effektivitet p\u00e5 lang sigt. <\/span><\/p>\n<p><span style=\"font-weight: 400;\">N\u00e5r udviklere har travlt med at f\u00e5 ML-systemer op at k\u00f8re, overser de m\u00e5ske de indviklede detaljer i datah\u00e5ndteringen eller faldgruberne ved at \"lime\" forskellige dele sammen. <\/span><\/p>\n<p><span style=\"font-weight: 400;\">Det fungerer m\u00e5ske p\u00e5 kort sigt, men kan f\u00f8re til et virvar, der er sv\u00e6rt at dissekere, opdatere eller forst\u00e5 senere.<\/span><\/p>\n<blockquote class=\"twitter-tweet\">\n<p dir=\"ltr\" lang=\"en\">\u26a0\ufe0f \u26a0\ufe0f \u26a0\ufe0f \u26a0\ufe0f \u26a0\ufe0f \u26a0\ufe0f \u26a0\ufe0f<\/p>\n<p>GenAI er en lavine af teknisk g\u00e6ld*, der venter p\u00e5 at ske<\/p>\n<p>Bare i denne uge<br \/>\n\ud83d\udc49ChatGPT gik \"bers\u00e6rk\" med n\u00e6sten ingen reel forklaring<br \/>\n\ud83d\udc49Sora kan ikke konsekvent udlede, hvor mange ben en kat har<br \/>\n\ud83d\udc49Geminis mangfoldighedsindsats k\u00f8rte helt af sporet.... <a href=\"https:\/\/t.co\/qzrVlpX9yz\">pic.twitter.com\/qzrVlpX9yz<\/a><\/p>\n<p>- Gary Marcus @ AAAI 2024 (@GaryMarcus) <a href=\"https:\/\/twitter.com\/GaryMarcus\/status\/1761414330577539340?ref_src=twsrc%5Etfw\">24. februar 2024<\/a><\/p><\/blockquote>\n<p><script async src=\"https:\/\/platform.twitter.com\/widgets.js\" charset=\"utf-8\"><\/script><br \/>\n<span style=\"font-weight: 400;\">For eksempel virker det effektivt at bruge ML-modeller som de er fra et bibliotek, indtil man sidder med et \"limkode\"-mareridt, hvor det meste af systemet bare er gaffatape, der holder sammen p\u00e5 stumper og stykker, som ikke var beregnet til at passe sammen.\u00a0<\/span><\/p>\n<p>Eller t\u00e6nk p\u00e5 \"r\u00f8rledningsjungler\", beskrevet i en <a href=\"https:\/\/proceedings.neurips.cc\/paper_files\/paper\/2015\/file\/86df7dcfd896fcaf2674f757a2463eba-Paper.pdf\" target=\"_blank\" rel=\"noopener\">tidligere artikel af D. Sculley<\/a> og kolleger, hvor dataforberedelse bliver en labyrint af sammenflettede processer, s\u00e5 det at foretage en \u00e6ndring f\u00f8les som at desarmere en bombe.<\/p>\n<h3>Konsekvenserne af teknisk g\u00e6ld<\/h3>\n<p><span style=\"font-weight: 400;\">Jo mere indviklet et system bliver, jo sv\u00e6rere er det at forbedre eller vedligeholde det. Det kv\u00e6ler ikke kun innovation, men kan ogs\u00e5 f\u00f8re til mere uhyggelige problemer.\u00a0<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Hvis et ML-system for eksempel begynder at tr\u00e6ffe beslutninger baseret p\u00e5 for\u00e6ldede eller forudindtagede data, fordi det er for besv\u00e6rligt at opdatere, kan det forst\u00e6rke eller <\/span><a href=\"https:\/\/dailyai.com\/da\/2023\/07\/unmasking-the-deep-seated-biases-in-ai-systems\/\"><span style=\"font-weight: 400;\">forst\u00e6rker samfundsm\u00e6ssige fordomme<\/span><\/a><span style=\"font-weight: 400;\">.\u00a0<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Desuden er der i kritiske applikationer som <\/span><a href=\"https:\/\/dailyai.com\/da\/2024\/02\/does-ai-display-racial-and-gender-bias-when-evaluating-images\/\"><span style=\"font-weight: 400;\">sundhedspleje<\/span><\/a><span style=\"font-weight: 400;\"> eller autonome k\u00f8ret\u00f8jer, kan en s\u00e5dan teknisk g\u00e6ld f\u00e5 alvorlige konsekvenser, ikke kun i form af tid og penge, men ogs\u00e5 for menneskers velbefindende.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Som unders\u00f8gelsen beskriver: \"Ikke al g\u00e6ld er n\u00f8dvendigvis d\u00e5rlig, men teknisk g\u00e6ld har en tendens til at vokse. Hvis man udskyder arbejdet for at betale den tilbage, resulterer det i stigende omkostninger, et skr\u00f8beligt system og mindre innovation.\"<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Det er ogs\u00e5 en p\u00e5mindelse til virksomheder og forbrugere om at kr\u00e6ve gennemsigtighed og ansvarlighed i de AI-teknologier, de tager i brug. <\/span><\/p>\n<p><span style=\"font-weight: 400;\">N\u00e5r alt kommer til alt, er m\u00e5let at udnytte kraften i AI til at g\u00f8re livet bedre, ikke at blive fanget i en endel\u00f8s cyklus af tilbagebetaling af teknisk g\u00e6ld.<\/span><\/p>\n<h2>Studie 2: Du kan ikke adskille hallucinationer fra LLM'er<\/h2>\n<p>I en anden, men <a href=\"https:\/\/arxiv.org\/pdf\/2401.11817.pdf\" target=\"_blank\" rel=\"noopener\">relateret unders\u00f8gelse<\/a> Fra National University of Singapore unders\u00f8gte forskerne Ziwei Xu, Sanjay Jain og Mohan Kankanhalli de iboende begr\u00e6nsninger i LLM'er.<\/p>\n<p><span style=\"font-weight: 400;\">\"Hallucinationer er uundg\u00e5elige: An Innate Limitation of Large Language Models\" udforsker arten af AI-hallucinationer, som beskriver tilf\u00e6lde, hvor AI-systemer genererer plausible, men un\u00f8jagtige eller helt fabrikerede oplysninger.\u00a0<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Hallucinationsf\u00e6nomenerne udg\u00f8r en stor teknisk udfordring, da de fremh\u00e6ver en grundl\u00e6ggende kl\u00f8ft mellem en AI-models output og det, der betragtes som \"grundsandheden\" - en ideel model, der altid producerer korrekt og logisk information.\u00a0<\/span><\/p>\n<p>At forst\u00e5, hvordan og hvorfor generativ AI hallucinerer, er altafg\u00f8rende, n\u00e5r teknologien integreres i kritiske sektorer som politi og retsv\u00e6sen, sundhedspleje og juridiske forhold.<\/p>\n<blockquote class=\"twitter-tweet\">\n<p dir=\"ltr\" lang=\"en\">Hvad nu, hvis man kunne *bevise*, at hallucinationer er uundg\u00e5elige hos LLM'er?<\/p>\n<p>Ville det \u00e6ndre sig?<br \/>\n- Hvordan ser du p\u00e5 LLM'er?<br \/>\n- Hvor meget vil du investere i dem?<br \/>\n- Hvor meget vil du prioritere forskning i alternativer?<\/p>\n<p>Det viser en ny artikel: <a href=\"https:\/\/t.co\/r0eP3mFxQg\">https:\/\/t.co\/r0eP3mFxQg<\/a><br \/>\nh\/t... <a href=\"https:\/\/t.co\/Id2kdaCSGk\">pic.twitter.com\/Id2kdaCSGk<\/a><\/p>\n<p>- Gary Marcus @ AAAI 2024 (@GaryMarcus) <a href=\"https:\/\/twitter.com\/GaryMarcus\/status\/1761764524674457662?ref_src=twsrc%5Etfw\">25. februar 2024<\/a><\/p><\/blockquote>\n<p><script async src=\"https:\/\/platform.twitter.com\/widgets.js\" charset=\"utf-8\"><\/script><\/p>\n<h3>Teoretisk grundlag for hallucinationer<\/h3>\n<p><span style=\"font-weight: 400;\">Unders\u00f8gelsen begynder med at opstille en teoretisk ramme for at forst\u00e5 hallucinationer hos LLM'er.\u00a0<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Forskere c<\/span><span style=\"font-weight: 400;\">reated en teoretisk model kendt som den \"formelle verden\". Dette forenklede, kontrollerede milj\u00f8 gjorde det muligt for dem at observere de forhold, hvorunder AI-modeller ikke stemmer overens med den grundl\u00e6ggende sandhed.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Derefter testede de to store familier af LLM'er:<\/span><\/p>\n<ol>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><b>Lama 2<\/b><span style=\"font-weight: 400;\">: Specifikt blev versionen med 70 milliarder parametre (llama2-70b-chat-hf), der er tilg\u00e6ngelig p\u00e5 HuggingFace, brugt. Denne model repr\u00e6senterer en af de nyere indgange til arenaen for store sprogmodeller, der er designet til en bred vifte af tekstgenererings- og forst\u00e5elsesopgaver.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><b>Generative fortr\u00e6net transformatorer (GPT)<\/b><span style=\"font-weight: 400;\">: Unders\u00f8gelsen omfattede test af GPT-3.5, specifikt den 175 milliarder parametre store gpt-3.5-turbo-16k-model, og GPT-4 (gpt-4-0613), hvor det n\u00f8jagtige antal parametre ikke er oplyst.\u00a0<\/span><\/li>\n<\/ol>\n<p><span style=\"font-weight: 400;\">LLM'erne blev bedt om at opremse strenge af en given l\u00e6ngde ved hj\u00e6lp af et bestemt alfabet, en tilsyneladende simpel beregningsopgave.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Mere specifikt fik modellerne til opgave at generere alle mulige strenge af varierende l\u00e6ngde fra 1 til 7 ved hj\u00e6lp af alfabeter med to tegn (f.eks. {a, b}) og tre tegn (f.eks. {a, b, c}).<\/span><\/p>\n<p><span style=\"font-weight: 400;\"> Outputtene blev vurderet ud fra, om de indeholdt alle og kun strenge af den angivne l\u00e6ngde fra det givne alfabet.<\/span><\/p>\n<h3>Resultater<\/h3>\n<p><span style=\"font-weight: 400;\">Resultaterne viste en klar begr\u00e6nsning i modellernes evne til at udf\u00f8re opgaven korrekt, n\u00e5r kompleksiteten steg (dvs. n\u00e5r strengl\u00e6ngden eller alfabetets st\u00f8rrelse steg). Mere specifikt:<\/span><\/p>\n<ul>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">Modellerne klarede sig godt med kortere strenge og mindre alfabeter, men vaklede, da opgavens kompleksitet steg.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">Selv den avancerede GPT-4-model, den mest sofistikerede LLM, der findes lige nu, kunne ikke liste alle strenge ud over visse l\u00e6ngder.<\/span><\/li>\n<\/ul>\n<p>Det viser, at hallucinationer ikke er en simpel fejl, der kan lappes eller rettes - de er et grundl\u00e6ggende aspekt af, hvordan disse modeller forst\u00e5r og gengiver det menneskelige sprog.<\/p>\n<p>Som unders\u00f8gelsen beskriver, <span class=\"css-1qaijid r-bcqeeo r-qvutc0 r-poiln3\">\"LLM'ere kan ikke l\u00e6re alt <\/span><span class=\"css-1qaijid r-bcqeeo r-qvutc0 r-poiln3 r-b88u0q\">af<\/span><span class=\"css-1qaijid r-bcqeeo r-qvutc0 r-poiln3\"> de beregnelige funktioner og vil derfor altid hallucinere. Eftersom den formelle verden <\/span><span class=\"css-1qaijid r-bcqeeo r-qvutc0 r-poiln3 r-b88u0q\">er<\/span><span class=\"css-1qaijid r-bcqeeo r-qvutc0 r-poiln3\"> en del <\/span><span class=\"css-1qaijid r-bcqeeo r-qvutc0 r-poiln3 r-b88u0q\">af<\/span><span class=\"css-1qaijid r-bcqeeo r-qvutc0 r-poiln3\"> den virkelige verden, som <\/span><span class=\"css-1qaijid r-bcqeeo r-qvutc0 r-poiln3 r-b88u0q\">er<\/span><span class=\"css-1qaijid r-bcqeeo r-qvutc0 r-poiln3\"> meget mere kompliceret, hallucinationer er ogs\u00e5 <\/span><span class=\"css-1qaijid r-bcqeeo r-qvutc0 r-poiln3 r-b88u0q\">uundg\u00e5elig<\/span><span class=\"css-1qaijid r-bcqeeo r-qvutc0 r-poiln3\"> for LLM'ere i den virkelige verden.\"<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Konsekvenserne for ans\u00f8gninger, hvor der er meget p\u00e5 spil, er enorme. I sektorer som sundhed, finans eller jura, hvor n\u00f8jagtigheden af oplysninger kan have alvorlige konsekvenser, kan det f\u00f8re til alvorlige fejl at stole p\u00e5 en LLM uden en sikkerhedsforanstaltning til at filtrere disse hallucinationer fra.<\/span><\/p>\n<p>Denne unders\u00f8gelse fangede AI-eksperten Dr. Gary Marcus' og den fremtr\u00e6dende kognitive psykolog Dr. Steven Pinkers opm\u00e6rksomhed.<\/p>\n<blockquote class=\"twitter-tweet\">\n<p dir=\"ltr\" lang=\"en\">Hallucinationer er uundg\u00e5elige med store sprogmodeller p\u00e5 grund af deres design: ingen repr\u00e6sentation af fakta eller ting, kun statistiske interkorrelationer. Nyt bevis p\u00e5 \"en medf\u00f8dt begr\u00e6nsning\" ved LLM'er. <a href=\"https:\/\/t.co\/Hl1kqxJGXt\">https:\/\/t.co\/Hl1kqxJGXt<\/a><\/p>\n<p>- Steven Pinker (@sapinker) <a href=\"https:\/\/twitter.com\/sapinker\/status\/1761801185181200410?ref_src=twsrc%5Etfw\">25. februar 2024<\/a><\/p><\/blockquote>\n<p><script async src=\"https:\/\/platform.twitter.com\/widgets.js\" charset=\"utf-8\"><\/script><\/p>\n<h2>Dybere problemer er p\u00e5 spil<\/h2>\n<p><span style=\"font-weight: 400;\">Akkumuleringen af teknisk g\u00e6ld og de uundg\u00e5elige hallucinationer i LLM'er er symptomatiske for et dybere problem - det nuv\u00e6rende paradigme for AI-udvikling kan i sagens natur v\u00e6re forkert indrettet til at skabe meget intelligente systemer og p\u00e5lideligt afstemt med menneskelige v\u00e6rdier og faktuel sandhed.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">P\u00e5 f\u00f8lsomme omr\u00e5der er det ikke nok at have et AI-system, der har ret det meste af tiden. Teknisk g\u00e6ld og hallucinationer truer begge modellens integritet over tid.\u00a0<\/span><\/p>\n<p>At l\u00f8se dette er ikke kun en teknisk udfordring, men en tv\u00e6rfaglig udfordring, der kr\u00e6ver input fra AI-etik, politik og dom\u00e6nespecifik ekspertise for at navigere sikkert.<\/p>\n<p>Lige nu er det tilsyneladende i modstrid med principperne i en branche, der lever op til mottoet \"move fast and break things\".<\/p>\n<p>Lad os h\u00e5be, at mennesker ikke er 'tingene'.<\/p>","protected":false},"excerpt":{"rendered":"<p>Efterh\u00e5nden som AI-systemer som store sprogmodeller (LLM'er) vokser i st\u00f8rrelse og kompleksitet, afd\u00e6kker forskere sp\u00e6ndende grundl\u00e6ggende begr\u00e6nsninger.  Nylige unders\u00f8gelser fra Google og University of Singapore har afd\u00e6kket mekanikken bag AI-\"hallucinationer\" - hvor modeller genererer overbevisende, men fabrikerede oplysninger - og ophobningen af \"teknisk g\u00e6ld\", som kan skabe rodede, up\u00e5lidelige systemer over tid. Ud over de tekniske udfordringer er det stadig et \u00e5bent sp\u00f8rgsm\u00e5l at tilpasse AI's evner og incitamenter til menneskelige v\u00e6rdier. N\u00e5r virksomheder som OpenAI skubber p\u00e5 mod kunstig generel intelligens (AGI), betyder det at sikre vejen frem at anerkende gr\u00e6nserne for de nuv\u00e6rende systemer. Men en omhyggelig anerkendelse af risici er<\/p>","protected":false},"author":2,"featured_media":10364,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[84],"tags":[480,118],"class_list":["post-10342","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-industry","tag-hallucinations","tag-llms"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v27.4 - https:\/\/yoast.com\/product\/yoast-seo-wordpress\/ -->\n<title>Generative AI systems, hallucinations, and mounting technical debt | DailyAI<\/title>\n<meta name=\"description\" content=\"As AI systems like large language models (LLMs) grow in size and complexity, researchers are uncovering intriguing fundamental limitations.\u00a0\" \/>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/dailyai.com\/da\/2024\/02\/generative-ai-systems-hallucinations-and-mounting-technical-debt\/\" \/>\n<meta property=\"og:locale\" content=\"da_DK\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Generative AI systems, hallucinations, and mounting technical debt | DailyAI\" \/>\n<meta property=\"og:description\" content=\"As AI systems like large language models (LLMs) grow in size and complexity, researchers are uncovering intriguing fundamental limitations.\u00a0\" \/>\n<meta property=\"og:url\" content=\"https:\/\/dailyai.com\/da\/2024\/02\/generative-ai-systems-hallucinations-and-mounting-technical-debt\/\" \/>\n<meta property=\"og:site_name\" content=\"DailyAI\" \/>\n<meta property=\"article:published_time\" content=\"2024-02-27T18:58:26+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2024-03-28T00:37:55+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/dailyai.com\/wp-content\/uploads\/2024\/02\/shutterstock_2146674289.jpg\" \/>\n\t<meta property=\"og:image:width\" content=\"1000\" \/>\n\t<meta property=\"og:image:height\" content=\"667\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/jpeg\" \/>\n<meta name=\"author\" content=\"Sam Jeans\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:creator\" content=\"@DailyAIOfficial\" \/>\n<meta name=\"twitter:site\" content=\"@DailyAIOfficial\" \/>\n<meta name=\"twitter:label1\" content=\"Skrevet af\" \/>\n\t<meta name=\"twitter:data1\" content=\"Sam Jeans\" \/>\n\t<meta name=\"twitter:label2\" content=\"Estimeret l\u00e6setid\" \/>\n\t<meta name=\"twitter:data2\" content=\"7 minutter\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"NewsArticle\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/02\\\/generative-ai-systems-hallucinations-and-mounting-technical-debt\\\/#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/02\\\/generative-ai-systems-hallucinations-and-mounting-technical-debt\\\/\"},\"author\":{\"name\":\"Sam Jeans\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/#\\\/schema\\\/person\\\/711e81f945549438e8bbc579efdeb3c9\"},\"headline\":\"Generative AI systems, hallucinations, and mounting technical debt\",\"datePublished\":\"2024-02-27T18:58:26+00:00\",\"dateModified\":\"2024-03-28T00:37:55+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/02\\\/generative-ai-systems-hallucinations-and-mounting-technical-debt\\\/\"},\"wordCount\":1569,\"publisher\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/#organization\"},\"image\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/02\\\/generative-ai-systems-hallucinations-and-mounting-technical-debt\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/dailyai.com\\\/wp-content\\\/uploads\\\/2024\\\/02\\\/shutterstock_2146674289.jpg\",\"keywords\":[\"Hallucinations\",\"LLMS\"],\"articleSection\":{\"1\":\"Industry\"},\"inLanguage\":\"da-DK\"},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/02\\\/generative-ai-systems-hallucinations-and-mounting-technical-debt\\\/\",\"url\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/02\\\/generative-ai-systems-hallucinations-and-mounting-technical-debt\\\/\",\"name\":\"Generative AI systems, hallucinations, and mounting technical debt | DailyAI\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/02\\\/generative-ai-systems-hallucinations-and-mounting-technical-debt\\\/#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/02\\\/generative-ai-systems-hallucinations-and-mounting-technical-debt\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/dailyai.com\\\/wp-content\\\/uploads\\\/2024\\\/02\\\/shutterstock_2146674289.jpg\",\"datePublished\":\"2024-02-27T18:58:26+00:00\",\"dateModified\":\"2024-03-28T00:37:55+00:00\",\"description\":\"As AI systems like large language models (LLMs) grow in size and complexity, researchers are uncovering intriguing fundamental limitations.\u00a0\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/02\\\/generative-ai-systems-hallucinations-and-mounting-technical-debt\\\/#breadcrumb\"},\"inLanguage\":\"da-DK\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/dailyai.com\\\/2024\\\/02\\\/generative-ai-systems-hallucinations-and-mounting-technical-debt\\\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"da-DK\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/02\\\/generative-ai-systems-hallucinations-and-mounting-technical-debt\\\/#primaryimage\",\"url\":\"https:\\\/\\\/dailyai.com\\\/wp-content\\\/uploads\\\/2024\\\/02\\\/shutterstock_2146674289.jpg\",\"contentUrl\":\"https:\\\/\\\/dailyai.com\\\/wp-content\\\/uploads\\\/2024\\\/02\\\/shutterstock_2146674289.jpg\",\"width\":1000,\"height\":667,\"caption\":\"technical debt AI\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/02\\\/generative-ai-systems-hallucinations-and-mounting-technical-debt\\\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\\\/\\\/dailyai.com\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Generative AI systems, hallucinations, and mounting technical debt\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/#website\",\"url\":\"https:\\\/\\\/dailyai.com\\\/\",\"name\":\"DailyAI\",\"description\":\"Your Daily Dose of AI News\",\"publisher\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/#organization\"},\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/dailyai.com\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"da-DK\"},{\"@type\":\"Organization\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/#organization\",\"name\":\"DailyAI\",\"url\":\"https:\\\/\\\/dailyai.com\\\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"da-DK\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/#\\\/schema\\\/logo\\\/image\\\/\",\"url\":\"https:\\\/\\\/dailyai.com\\\/wp-content\\\/uploads\\\/2023\\\/06\\\/Daily-Ai_TL_colour.png\",\"contentUrl\":\"https:\\\/\\\/dailyai.com\\\/wp-content\\\/uploads\\\/2023\\\/06\\\/Daily-Ai_TL_colour.png\",\"width\":4501,\"height\":934,\"caption\":\"DailyAI\"},\"image\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/#\\\/schema\\\/logo\\\/image\\\/\"},\"sameAs\":[\"https:\\\/\\\/x.com\\\/DailyAIOfficial\",\"https:\\\/\\\/www.linkedin.com\\\/company\\\/dailyaiofficial\\\/\",\"https:\\\/\\\/www.youtube.com\\\/@DailyAIOfficial\"]},{\"@type\":\"Person\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/#\\\/schema\\\/person\\\/711e81f945549438e8bbc579efdeb3c9\",\"name\":\"Sam Jeans\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"da-DK\",\"@id\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/a24a4a8f8e2a1a275b7491dc9c9f032c401eabf23c3206da4628dc84b6dac5c8?s=96&d=robohash&r=g\",\"url\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/a24a4a8f8e2a1a275b7491dc9c9f032c401eabf23c3206da4628dc84b6dac5c8?s=96&d=robohash&r=g\",\"contentUrl\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/a24a4a8f8e2a1a275b7491dc9c9f032c401eabf23c3206da4628dc84b6dac5c8?s=96&d=robohash&r=g\",\"caption\":\"Sam Jeans\"},\"description\":\"Sam is a science and technology writer who has worked in various AI startups. When he\u2019s not writing, he can be found reading medical journals or digging through boxes of vinyl records.\",\"sameAs\":[\"https:\\\/\\\/www.linkedin.com\\\/in\\\/sam-jeans-6746b9142\\\/\"],\"url\":\"https:\\\/\\\/dailyai.com\\\/da\\\/author\\\/samjeans\\\/\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"Generative AI-systemer, hallucinationer og stigende teknisk g\u00e6ld | DailyAI","description":"Efterh\u00e5nden som AI-systemer som store sprogmodeller (LLM'er) vokser i st\u00f8rrelse og kompleksitet, afd\u00e6kker forskere sp\u00e6ndende grundl\u00e6ggende begr\u00e6nsninger.\u00a0","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/dailyai.com\/da\/2024\/02\/generative-ai-systems-hallucinations-and-mounting-technical-debt\/","og_locale":"da_DK","og_type":"article","og_title":"Generative AI systems, hallucinations, and mounting technical debt | DailyAI","og_description":"As AI systems like large language models (LLMs) grow in size and complexity, researchers are uncovering intriguing fundamental limitations.\u00a0","og_url":"https:\/\/dailyai.com\/da\/2024\/02\/generative-ai-systems-hallucinations-and-mounting-technical-debt\/","og_site_name":"DailyAI","article_published_time":"2024-02-27T18:58:26+00:00","article_modified_time":"2024-03-28T00:37:55+00:00","og_image":[{"width":1000,"height":667,"url":"https:\/\/dailyai.com\/wp-content\/uploads\/2024\/02\/shutterstock_2146674289.jpg","type":"image\/jpeg"}],"author":"Sam Jeans","twitter_card":"summary_large_image","twitter_creator":"@DailyAIOfficial","twitter_site":"@DailyAIOfficial","twitter_misc":{"Skrevet af":"Sam Jeans","Estimeret l\u00e6setid":"7 minutter"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"NewsArticle","@id":"https:\/\/dailyai.com\/2024\/02\/generative-ai-systems-hallucinations-and-mounting-technical-debt\/#article","isPartOf":{"@id":"https:\/\/dailyai.com\/2024\/02\/generative-ai-systems-hallucinations-and-mounting-technical-debt\/"},"author":{"name":"Sam Jeans","@id":"https:\/\/dailyai.com\/#\/schema\/person\/711e81f945549438e8bbc579efdeb3c9"},"headline":"Generative AI systems, hallucinations, and mounting technical debt","datePublished":"2024-02-27T18:58:26+00:00","dateModified":"2024-03-28T00:37:55+00:00","mainEntityOfPage":{"@id":"https:\/\/dailyai.com\/2024\/02\/generative-ai-systems-hallucinations-and-mounting-technical-debt\/"},"wordCount":1569,"publisher":{"@id":"https:\/\/dailyai.com\/#organization"},"image":{"@id":"https:\/\/dailyai.com\/2024\/02\/generative-ai-systems-hallucinations-and-mounting-technical-debt\/#primaryimage"},"thumbnailUrl":"https:\/\/dailyai.com\/wp-content\/uploads\/2024\/02\/shutterstock_2146674289.jpg","keywords":["Hallucinations","LLMS"],"articleSection":{"1":"Industry"},"inLanguage":"da-DK"},{"@type":"WebPage","@id":"https:\/\/dailyai.com\/2024\/02\/generative-ai-systems-hallucinations-and-mounting-technical-debt\/","url":"https:\/\/dailyai.com\/2024\/02\/generative-ai-systems-hallucinations-and-mounting-technical-debt\/","name":"Generative AI-systemer, hallucinationer og stigende teknisk g\u00e6ld | DailyAI","isPartOf":{"@id":"https:\/\/dailyai.com\/#website"},"primaryImageOfPage":{"@id":"https:\/\/dailyai.com\/2024\/02\/generative-ai-systems-hallucinations-and-mounting-technical-debt\/#primaryimage"},"image":{"@id":"https:\/\/dailyai.com\/2024\/02\/generative-ai-systems-hallucinations-and-mounting-technical-debt\/#primaryimage"},"thumbnailUrl":"https:\/\/dailyai.com\/wp-content\/uploads\/2024\/02\/shutterstock_2146674289.jpg","datePublished":"2024-02-27T18:58:26+00:00","dateModified":"2024-03-28T00:37:55+00:00","description":"Efterh\u00e5nden som AI-systemer som store sprogmodeller (LLM'er) vokser i st\u00f8rrelse og kompleksitet, afd\u00e6kker forskere sp\u00e6ndende grundl\u00e6ggende begr\u00e6nsninger.\u00a0","breadcrumb":{"@id":"https:\/\/dailyai.com\/2024\/02\/generative-ai-systems-hallucinations-and-mounting-technical-debt\/#breadcrumb"},"inLanguage":"da-DK","potentialAction":[{"@type":"ReadAction","target":["https:\/\/dailyai.com\/2024\/02\/generative-ai-systems-hallucinations-and-mounting-technical-debt\/"]}]},{"@type":"ImageObject","inLanguage":"da-DK","@id":"https:\/\/dailyai.com\/2024\/02\/generative-ai-systems-hallucinations-and-mounting-technical-debt\/#primaryimage","url":"https:\/\/dailyai.com\/wp-content\/uploads\/2024\/02\/shutterstock_2146674289.jpg","contentUrl":"https:\/\/dailyai.com\/wp-content\/uploads\/2024\/02\/shutterstock_2146674289.jpg","width":1000,"height":667,"caption":"technical debt AI"},{"@type":"BreadcrumbList","@id":"https:\/\/dailyai.com\/2024\/02\/generative-ai-systems-hallucinations-and-mounting-technical-debt\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/dailyai.com\/"},{"@type":"ListItem","position":2,"name":"Generative AI systems, hallucinations, and mounting technical debt"}]},{"@type":"WebSite","@id":"https:\/\/dailyai.com\/#website","url":"https:\/\/dailyai.com\/","name":"DailyAI","description":"Din daglige dosis af AI-nyheder","publisher":{"@id":"https:\/\/dailyai.com\/#organization"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/dailyai.com\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"da-DK"},{"@type":"Organization","@id":"https:\/\/dailyai.com\/#organization","name":"DailyAI","url":"https:\/\/dailyai.com\/","logo":{"@type":"ImageObject","inLanguage":"da-DK","@id":"https:\/\/dailyai.com\/#\/schema\/logo\/image\/","url":"https:\/\/dailyai.com\/wp-content\/uploads\/2023\/06\/Daily-Ai_TL_colour.png","contentUrl":"https:\/\/dailyai.com\/wp-content\/uploads\/2023\/06\/Daily-Ai_TL_colour.png","width":4501,"height":934,"caption":"DailyAI"},"image":{"@id":"https:\/\/dailyai.com\/#\/schema\/logo\/image\/"},"sameAs":["https:\/\/x.com\/DailyAIOfficial","https:\/\/www.linkedin.com\/company\/dailyaiofficial\/","https:\/\/www.youtube.com\/@DailyAIOfficial"]},{"@type":"Person","@id":"https:\/\/dailyai.com\/#\/schema\/person\/711e81f945549438e8bbc579efdeb3c9","name":"Sam Jeans","image":{"@type":"ImageObject","inLanguage":"da-DK","@id":"https:\/\/secure.gravatar.com\/avatar\/a24a4a8f8e2a1a275b7491dc9c9f032c401eabf23c3206da4628dc84b6dac5c8?s=96&d=robohash&r=g","url":"https:\/\/secure.gravatar.com\/avatar\/a24a4a8f8e2a1a275b7491dc9c9f032c401eabf23c3206da4628dc84b6dac5c8?s=96&d=robohash&r=g","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/a24a4a8f8e2a1a275b7491dc9c9f032c401eabf23c3206da4628dc84b6dac5c8?s=96&d=robohash&r=g","caption":"Sam Jeans"},"description":"Sam er videnskabs- og teknologiforfatter og har arbejdet i forskellige AI-startups. N\u00e5r han ikke skriver, kan han finde p\u00e5 at l\u00e6se medicinske tidsskrifter eller grave i kasser med vinylplader.","sameAs":["https:\/\/www.linkedin.com\/in\/sam-jeans-6746b9142\/"],"url":"https:\/\/dailyai.com\/da\/author\/samjeans\/"}]}},"_links":{"self":[{"href":"https:\/\/dailyai.com\/da\/wp-json\/wp\/v2\/posts\/10342","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/dailyai.com\/da\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/dailyai.com\/da\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/dailyai.com\/da\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/dailyai.com\/da\/wp-json\/wp\/v2\/comments?post=10342"}],"version-history":[{"count":11,"href":"https:\/\/dailyai.com\/da\/wp-json\/wp\/v2\/posts\/10342\/revisions"}],"predecessor-version":[{"id":10427,"href":"https:\/\/dailyai.com\/da\/wp-json\/wp\/v2\/posts\/10342\/revisions\/10427"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/dailyai.com\/da\/wp-json\/wp\/v2\/media\/10364"}],"wp:attachment":[{"href":"https:\/\/dailyai.com\/da\/wp-json\/wp\/v2\/media?parent=10342"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/dailyai.com\/da\/wp-json\/wp\/v2\/categories?post=10342"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/dailyai.com\/da\/wp-json\/wp\/v2\/tags?post=10342"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}