{"id":12519,"date":"2024-05-22T18:16:07","date_gmt":"2024-05-22T18:16:07","guid":{"rendered":"https:\/\/dailyai.com\/?p=12519"},"modified":"2024-05-24T10:16:44","modified_gmt":"2024-05-24T10:16:44","slug":"decoding-the-ai-mind-anthropic-researchers-peer-inside-the-black-box","status":"publish","type":"post","link":"https:\/\/dailyai.com\/sv\/2024\/05\/decoding-the-ai-mind-anthropic-researchers-peer-inside-the-black-box\/","title":{"rendered":"Avkodning av AI-sinnet: Antropiska forskare tittar in i den \"svarta l\u00e5dan\""},"content":{"rendered":"<p><b><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\">Anthropic<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span> forskare framg\u00e5ngsrikt identifierat miljontals begrepp inom <span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\">Claude<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span> Sonnet, en av deras avancerade LLM:er.<\/b><\/p>\n<p><span style=\"font-weight: 400;\">AI-modeller betraktas ofta som svarta l\u00e5dor, vilket inneb\u00e4r att man inte kan \"se\" in i dem f\u00f6r att f\u00f6rst\u00e5 exakt hur de fungerar.\u00a0<\/span><\/p>\n<p><span style=\"font-weight: 400;\">N\u00e4r du ger en LLM en inmatning genererar den ett svar, men resonemanget bakom dess val \u00e4r inte tydligt. <\/span><\/p>\n<p><span style=\"font-weight: 400;\">Din input g\u00e5r in och output kommer ut - och inte ens AI-utvecklarna sj\u00e4lva f\u00f6rst\u00e5r riktigt vad som h\u00e4nder inuti den d\u00e4r \"l\u00e5dan\".\u00a0<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Neurala n\u00e4tverk skapar sina egna interna representationer av information n\u00e4r de mappar inmatningar till utmatningar under datatr\u00e4ningen.\u00a0<\/span><span style=\"font-weight: 400;\">Byggstenarna i denna process, som kallas \"neuronaktiveringar\", representeras av numeriska v\u00e4rden. <\/span><\/p>\n<p><span style=\"font-weight: 400;\">Varje begrepp \u00e4r f\u00f6rdelat p\u00e5 flera neuroner, och varje neuron bidrar till att representera flera begrepp, vilket g\u00f6r det sv\u00e5rt att mappa begrepp direkt till enskilda neuroner.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Detta \u00e4r i stort sett analogt med v\u00e5ra m\u00e4nskliga hj\u00e4rnor. Precis som v\u00e5ra hj\u00e4rnor bearbetar sinnesintryck och genererar tankar, beteenden och minnen, \u00e4r de miljarder, till och med biljoner, processer som ligger bakom dessa funktioner i huvudsak ok\u00e4nda f\u00f6r vetenskapen.<\/span><\/p>\n<p><a href=\"https:\/\/www.anthropic.com\/research\/mapping-mind-language-model\"><span style=\"font-weight: 400;\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\">Anthropic<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>studie<\/span><\/a><span style=\"font-weight: 400;\"> f\u00f6rs\u00f6ker se in i AI:s svarta l\u00e5da med en teknik som kallas \"dictionary learning\".\u00a0<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Detta inneb\u00e4r att komplexa m\u00f6nster i en AI-modell bryts ned till linj\u00e4ra byggstenar eller \"atomer\" som \u00e4r intuitivt begripliga f\u00f6r m\u00e4nniskor.<\/span><\/p>\n<h2>Mappning av LLM:er med Dictionary Learning<\/h2>\n<p><span style=\"font-weight: 400;\">I oktober 2023, <span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\">Anthropic<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span> till\u00e4mpade denna metod p\u00e5 en liten \"leksaksspr\u00e5kmodell\" och fann sammanh\u00e4ngande funktioner som motsvarade begrepp som versaler, DNA-sekvenser, efternamn i citat, matematiska substantiv eller funktionsargument i Python-kod.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Den senaste studien skalar upp tekniken s\u00e5 att den fungerar f\u00f6r dagens st\u00f6rre AI-spr\u00e5kmodeller, i det h\u00e4r fallet, <span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\">Anthropic<\/span><\/span><\/span><\/span>'s <span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\">Claude<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span> 3 Sonett.\u00a0<\/span><\/p>\n<p><span style=\"font-weight: 400;\">H\u00e4r f\u00f6ljer en steg-f\u00f6r-stegbeskrivning av hur studien gick till:<\/span><\/p>\n<h4><b>Identifiera m\u00f6nster med hj\u00e4lp av ordboksinl\u00e4rning<\/b><\/h4>\n<p><span style=\"font-weight: 400;\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\">Anthropic<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span> anv\u00e4nde ordboksinl\u00e4rning f\u00f6r att analysera neuronaktiveringar i olika sammanhang och identifiera gemensamma m\u00f6nster. <\/span><\/p>\n<p><span style=\"font-weight: 400;\">Ordboksinl\u00e4rning grupperar dessa aktiveringar till en mindre upps\u00e4ttning meningsfulla \"funktioner\", som representerar begrepp p\u00e5 h\u00f6gre niv\u00e5 som modellen l\u00e4rt sig. <\/span><\/p>\n<p><span style=\"font-weight: 400;\">Genom att identifiera dessa egenskaper kan forskarna b\u00e4ttre f\u00f6rst\u00e5 hur modellen bearbetar och representerar information.<\/span><\/p>\n<h4><b>Extrahering av funktioner fr\u00e5n mellanlagret<\/b><\/h4>\n<p><span style=\"font-weight: 400;\">Forskarna fokuserade p\u00e5 det mellersta lagret av <span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\">Claude<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span> 3.0 Sonnet, som fungerar som en kritisk punkt i modellens bearbetningspipeline. <\/span><\/p>\n<p><span style=\"font-weight: 400;\">Genom att till\u00e4mpa ordboksinl\u00e4rning p\u00e5 detta lager extraheras miljontals funktioner som f\u00e5ngar modellens interna representationer och inl\u00e4rda begrepp i detta skede. <\/span><\/p>\n<p><span style=\"font-weight: 400;\">Genom att extrahera funktioner fr\u00e5n det mellersta lagret kan forskare unders\u00f6ka modellens f\u00f6rst\u00e5else av information <em>efter<\/em> den har bearbetat inmatningen <em>f\u00f6re<\/em> generera den slutliga produktionen.<\/span><\/p>\n<h4><b>Uppt\u00e4cka olika och abstrakta begrepp<\/b><\/h4>\n<p><span style=\"font-weight: 400;\">De extraherade funktionerna avsl\u00f6jade ett expansivt utbud av begrepp som l\u00e4rts in av <span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\">Claude<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>fr\u00e5n konkreta enheter som st\u00e4der och m\u00e4nniskor till abstrakta begrepp relaterade till vetenskapliga omr\u00e5den och programmeringssyntax. <\/span><\/p>\n<p><span style=\"font-weight: 400;\">Intressant nog visade det sig att funktionerna var multimodala och svarade p\u00e5 b\u00e5de textuella och visuella intryck, vilket tyder p\u00e5 att modellen kan l\u00e4ra sig och representera begrepp i olika modaliteter. <\/span><\/p>\n<p><span style=\"font-weight: 400;\">Dessutom tyder de flerspr\u00e5kiga funktionerna p\u00e5 att modellen kan f\u00f6rst\u00e5 begrepp som uttrycks p\u00e5 olika spr\u00e5k.<\/span><\/p>\n<p><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"598\" class=\"wp-image-12521 size-large\" src=\"https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/c896a301ad3d1ef4237cb05b68d78b467c444097-2200x1284-1-1024x598.webp\" alt=\"&lt;span class=\" srcset=\"https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/c896a301ad3d1ef4237cb05b68d78b467c444097-2200x1284-1-1024x598.webp 1024w, https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/c896a301ad3d1ef4237cb05b68d78b467c444097-2200x1284-1-300x175.webp 300w, https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/c896a301ad3d1ef4237cb05b68d78b467c444097-2200x1284-1-768x448.webp 768w, https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/c896a301ad3d1ef4237cb05b68d78b467c444097-2200x1284-1-1536x896.webp 1536w, https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/c896a301ad3d1ef4237cb05b68d78b467c444097-2200x1284-1-2048x1195.webp 2048w, https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/c896a301ad3d1ef4237cb05b68d78b467c444097-2200x1284-1-18x12.webp 18w, https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/c896a301ad3d1ef4237cb05b68d78b467c444097-2200x1284-1-60x35.webp 60w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/p>\n<h3><b>Analys av begreppens organisation<\/b><\/h3>\n<p><span style=\"font-weight: 400;\">F\u00f6r att f\u00f6rst\u00e5 hur modellen organiserar och relaterar olika begrepp analyserade forskarna likheten mellan funktioner baserat p\u00e5 deras aktiveringsm\u00f6nster. <\/span><\/p>\n<p><span style=\"font-weight: 400;\">De uppt\u00e4ckte att funktioner som representerar relaterade begrepp tenderade att gruppera sig tillsammans. Till exempel uppvisade funktioner som associerades med st\u00e4der eller vetenskapliga discipliner h\u00f6gre likhet med varandra \u00e4n med funktioner som representerade orelaterade koncept. <\/span><\/p>\n<p><span style=\"font-weight: 400;\">Detta tyder p\u00e5 att modellens interna organisation av begrepp i viss m\u00e5n \u00f6verensst\u00e4mmer med m\u00e4nskliga intuitioner om begreppsm\u00e4ssiga relationer.<\/span><\/p>\n<figure id=\"attachment_12520\" aria-describedby=\"caption-attachment-12520\" style=\"width: 1024px\" class=\"wp-caption aligncenter\"><img loading=\"lazy\" decoding=\"async\" class=\"wp-image-12520 size-large\" src=\"https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/fe4d42c004bf43efda0f5921adfedd2f8f42e417-2200x2140-1-1024x996.webp\" alt=\"antropisk\" width=\"1024\" height=\"996\" srcset=\"https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/fe4d42c004bf43efda0f5921adfedd2f8f42e417-2200x2140-1-1024x996.webp 1024w, https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/fe4d42c004bf43efda0f5921adfedd2f8f42e417-2200x2140-1-300x292.webp 300w, https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/fe4d42c004bf43efda0f5921adfedd2f8f42e417-2200x2140-1-768x747.webp 768w, https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/fe4d42c004bf43efda0f5921adfedd2f8f42e417-2200x2140-1-1536x1494.webp 1536w, https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/fe4d42c004bf43efda0f5921adfedd2f8f42e417-2200x2140-1-2048x1992.webp 2048w, https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/fe4d42c004bf43efda0f5921adfedd2f8f42e417-2200x2140-1-12x12.webp 12w, https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/fe4d42c004bf43efda0f5921adfedd2f8f42e417-2200x2140-1-60x58.webp 60w, https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/fe4d42c004bf43efda0f5921adfedd2f8f42e417-2200x2140-1-24x24.webp 24w, https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/fe4d42c004bf43efda0f5921adfedd2f8f42e417-2200x2140-1-48x48.webp 48w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><figcaption id=\"caption-attachment-12520\" class=\"wp-caption-text\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\">Anthropic<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span> klarade av att kartl\u00e4gga abstrakta begrepp som \"inre konflikt\". K\u00e4lla: <a href=\"https:\/\/www.anthropic.com\/research\/mapping-mind-language-model\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\">Anthropic<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/a>.<\/figcaption><\/figure>\n<h4><b>Verifiering av funktionerna<\/b><\/h4>\n<p><span style=\"font-weight: 400;\">F\u00f6r att bekr\u00e4fta att de identifierade funktionerna direkt p\u00e5verkar modellens beteende och utdata genomf\u00f6rde forskarna experiment med \"funktionsstyrning\". <\/span><\/p>\n<p><span style=\"font-weight: 400;\">Detta innebar att man selektivt f\u00f6rst\u00e4rkte eller undertryckte aktiveringen av specifika funktioner under modellens bearbetning och observerade hur detta p\u00e5verkade dess svar. <\/span><\/p>\n<p><span style=\"font-weight: 400;\">Genom att manipulera enskilda funktioner kunde forskarna etablera en direkt koppling mellan enskilda funktioner och modellens beteende. Till exempel, genom att f\u00f6rst\u00e4rka en funktion relaterad till en specifik stad fick modellen att generera stadsbaserade utg\u00e5ngar, \u00e4ven i irrelevanta sammanhang.<\/span><\/p>\n<p>L\u00e4s hela studien <a href=\"https:\/\/transformer-circuits.pub\/2024\/scaling-monosemanticity\/index.html\" target=\"_blank\" rel=\"noopener\">h\u00e4r<\/a>.<\/p>\n<p><iframe loading=\"lazy\" title=\"Ordboksinl\u00e4rning p\u00e5 Claude 3 Sonnet\" width=\"1080\" height=\"608\" src=\"https:\/\/www.youtube.com\/embed\/CJIbCV92d88?feature=oembed\" frameborder=\"0\" allow=\"accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share\" referrerpolicy=\"strict-origin-when-cross-origin\" allowfullscreen><\/iframe><\/p>\n<h2>Varf\u00f6r tolkningsbarhet \u00e4r avg\u00f6rande f\u00f6r AI-s\u00e4kerhet<\/h2>\n<p><span style=\"font-weight: 400;\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\">Anthropic<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>'s forskning \u00e4r i grunden relevant f\u00f6r AI-tolkningsbarhet och, i f\u00f6rl\u00e4ngningen, s\u00e4kerhet.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Att f\u00f6rst\u00e5 hur LLM:er bearbetar och representerar information hj\u00e4lper forskare att f\u00f6rst\u00e5 och minska riskerna. Det <\/span><span style=\"font-weight: 400;\">l\u00e4gger grunden f\u00f6r att utveckla mer transparenta och f\u00f6rklarliga AI-system.\u00a0<\/span><\/p>\n<p>Som <span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\">Anthropic<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span> f\u00f6rklarar: \"Vi hoppas att vi och andra kan anv\u00e4nda dessa uppt\u00e4ckter f\u00f6r att g\u00f6ra modeller s\u00e4krare. Det kan till exempel vara m\u00f6jligt att anv\u00e4nda de tekniker som beskrivs h\u00e4r f\u00f6r att \u00f6vervaka AI-system f\u00f6r vissa farliga beteenden (som att lura anv\u00e4ndaren), f\u00f6r att styra dem mot \u00f6nskv\u00e4rda resultat (debiasing) eller f\u00f6r att helt ta bort vissa farliga \u00e4mnen.\"<\/p>\n<p><span style=\"font-weight: 400;\">Att f\u00e5 en st\u00f6rre f\u00f6rst\u00e5else f\u00f6r AI:s beteende blir allt viktigare n\u00e4r de blir allest\u00e4des n\u00e4rvarande i kritiska beslutsprocesser inom omr\u00e5den som sjukv\u00e5rd, finans och straffr\u00e4tt. Det hj\u00e4lper ocks\u00e5 till att avsl\u00f6ja grundorsaken till <a href=\"https:\/\/dailyai.com\/sv\/2023\/06\/anthropic-releases-paper-highlighting-the-potential-bias-of-large-language-models\/\">f\u00f6rsp\u00e4nning<\/a>, hallucinationer och andra o\u00f6nskade eller of\u00f6ruts\u00e4gbara beteenden.\u00a0<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Till exempel kan en <\/span><a href=\"https:\/\/dailyai.com\/sv\/2024\/01\/researchers-aim-to-determine-how-black-box-drug-discovery-models-work\/\"><span style=\"font-weight: 400;\">nyligen genomf\u00f6rd studie<\/span><\/a><span style=\"font-weight: 400;\"> fr\u00e5n University of Bonn avsl\u00f6jade hur grafneurala n\u00e4tverk (GNN) som anv\u00e4nds f\u00f6r l\u00e4kemedelsuppt\u00e4ckt \u00e4r starkt beroende av att \u00e5terkalla likheter fr\u00e5n tr\u00e4ningsdata snarare \u00e4n att verkligen l\u00e4ra sig komplexa nya kemiska interaktioner. <\/span><\/p>\n<p><span style=\"font-weight: 400;\">Detta g\u00f6r det sv\u00e5rt att f\u00f6rst\u00e5 exakt hur dessa modeller best\u00e4mmer nya intressanta f\u00f6reningar.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">F\u00f6rra \u00e5ret var <\/span><a href=\"https:\/\/dailyai.com\/sv\/2023\/09\/the-uk-government-wants-to-see-inside-ais-black-box\/\"><span style=\"font-weight: 400;\">Den brittiska regeringen f\u00f6rhandlade<\/span><\/a><span style=\"font-weight: 400;\"> med stora teknikj\u00e4ttar som <span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\">OpenAI<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span> och <span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\">DeepMind<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>som vill ha tillg\u00e5ng till sina AI-systems interna beslutsprocesser.\u00a0<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Reglering som den <a href=\"https:\/\/dailyai.com\/sv\/2024\/03\/the-eu-ai-act-passed-in-a-landslide-and-will-come-into-force-this-year\/\">EU:s AI-lag<\/a> kommer att pressa AI-f\u00f6retagen att bli mer transparenta, \u00e4ven om aff\u00e4rshemligheter verkar vara s\u00e4kra p\u00e5 att f\u00f6rbli inl\u00e5sta.\u00a0<\/span><\/p>\n<p><span style=\"font-weight: 400;\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\">Anthropic<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>'s forskning ger en glimt av vad som finns inuti l\u00e5dan genom att \"mappa\" information \u00f6ver modellen.\u00a0<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Sanningen \u00e4r dock att dessa modeller \u00e4r s\u00e5 omfattande att de genom <span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\"><span class=\"noTranslate\" data-no-translation=\"\">Anthropic<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\"Vi tror att det \u00e4r ganska troligt att vi \u00e4r storleksordningar korta, och att om vi ville f\u00e5 alla funktioner - i alla lager! - skulle vi beh\u00f6va anv\u00e4nda mycket mer databehandling \u00e4n den totala databehandling som beh\u00f6vs f\u00f6r att tr\u00e4na de underliggande modellerna.\"<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Det \u00e4r en intressant po\u00e4ng - att omv\u00e4nt konstruera en modell \u00e4r mer ber\u00e4kningsm\u00e4ssigt komplext \u00e4n att konstruera modellen i f\u00f6rsta hand.\u00a0<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Det p\u00e5minner om enormt dyra neurovetenskapliga projekt som <a href=\"https:\/\/dailyai.com\/sv\/2023\/08\/exploring-impact-of-the-human-brain-project-hbp-on-science-and-ai\/\">Human Brain Project (HBP)<\/a>som satsade miljarder p\u00e5 att kartl\u00e4gga v\u00e5ra egna m\u00e4nskliga hj\u00e4rnor, men misslyckades i slut\u00e4ndan.\u00a0<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Underskatta aldrig hur mycket som finns i den svarta l\u00e5dan.\u00a0<\/span><\/p>","protected":false},"excerpt":{"rendered":"<p>Antropiska forskare har lyckats identifiera miljontals begrepp i Claude Sonnet, en av deras avancerade LLM:er. AI-modeller betraktas ofta som svarta l\u00e5dor, vilket inneb\u00e4r att man inte kan \"se\" in i dem f\u00f6r att f\u00f6rst\u00e5 exakt hur de fungerar.  N\u00e4r du f\u00f6rser en LLM med en inmatning genererar den ett svar, men resonemanget bakom dess val \u00e4r inte tydligt. Inmatningen g\u00e5r in och utmatningen kommer ut - och inte ens AI-utvecklarna sj\u00e4lva f\u00f6rst\u00e5r riktigt vad som h\u00e4nder inuti den d\u00e4r \"l\u00e5dan\".  Neurala n\u00e4tverk skapar sina egna interna representationer av information n\u00e4r de mappar inmatningar till utmatningar under datatr\u00e4ningen. Byggstenarna i detta<\/p>","protected":false},"author":2,"featured_media":12522,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[84],"tags":[148,105],"class_list":["post-12519","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-industry","tag-anthropic","tag-machine-learning"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v27.4 - https:\/\/yoast.com\/product\/yoast-seo-wordpress\/ -->\n<title>Decoding the AI mind: Anthropic researchers peer inside the &quot;black box&quot; | DailyAI<\/title>\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\/sv\/2024\/05\/decoding-the-ai-mind-anthropic-researchers-peer-inside-the-black-box\/\" \/>\n<meta property=\"og:locale\" content=\"sv_SE\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Decoding the AI mind: Anthropic researchers peer inside the &quot;black box&quot; | DailyAI\" \/>\n<meta property=\"og:description\" content=\"Anthropic researchers successfully identified millions of concepts within Claude Sonnet, one of their advanced LLMs. AI models are often considered black boxes, meaning you can&#8217;t &#8216;see&#8217; inside them to understand exactly how they work.\u00a0 When you provide an LLM with an input, it generates a response, but the reasoning behind its choices isn\u2019t clear. Your input goes in, and the output comes out \u2013 and even the AI developers themselves don&#8217;t truly understand what happens inside that &#8216;box.&#8217;\u00a0 Neural networks create their own internal representations of information when they map inputs to outputs during data training.\u00a0The building blocks of this\" \/>\n<meta property=\"og:url\" content=\"https:\/\/dailyai.com\/sv\/2024\/05\/decoding-the-ai-mind-anthropic-researchers-peer-inside-the-black-box\/\" \/>\n<meta property=\"og:site_name\" content=\"DailyAI\" \/>\n<meta property=\"article:published_time\" content=\"2024-05-22T18:16:07+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2024-05-24T10:16:44+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/DALL\u00b7E-2024-05-22-19.15.50-Anthropic-researchers-studying-artificial-intelligence-inside-a-high-tech-laboratory-with-a-prominent-black-box-in-the-center.-The-scene-features-adva.webp\" \/>\n\t<meta property=\"og:image:width\" content=\"1792\" \/>\n\t<meta property=\"og:image:height\" content=\"1024\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/webp\" \/>\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=\"Skriven av\" \/>\n\t<meta name=\"twitter:data1\" content=\"Sam Jeans\" \/>\n\t<meta name=\"twitter:label2\" content=\"Ber\u00e4knad l\u00e4stid\" \/>\n\t<meta name=\"twitter:data2\" content=\"6 minuter\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"NewsArticle\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/05\\\/decoding-the-ai-mind-anthropic-researchers-peer-inside-the-black-box\\\/#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/05\\\/decoding-the-ai-mind-anthropic-researchers-peer-inside-the-black-box\\\/\"},\"author\":{\"name\":\"Sam Jeans\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/#\\\/schema\\\/person\\\/711e81f945549438e8bbc579efdeb3c9\"},\"headline\":\"Decoding the AI mind: Anthropic researchers peer inside the &#8220;black box&#8221;\",\"datePublished\":\"2024-05-22T18:16:07+00:00\",\"dateModified\":\"2024-05-24T10:16:44+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/05\\\/decoding-the-ai-mind-anthropic-researchers-peer-inside-the-black-box\\\/\"},\"wordCount\":1075,\"publisher\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/#organization\"},\"image\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/05\\\/decoding-the-ai-mind-anthropic-researchers-peer-inside-the-black-box\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/dailyai.com\\\/wp-content\\\/uploads\\\/2024\\\/05\\\/DALL\u00b7E-2024-05-22-19.15.50-Anthropic-researchers-studying-artificial-intelligence-inside-a-high-tech-laboratory-with-a-prominent-black-box-in-the-center.-The-scene-features-adva.webp\",\"keywords\":[\"Anthropic\",\"machine learning\"],\"articleSection\":[\"Industry\"],\"inLanguage\":\"sv-SE\"},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/05\\\/decoding-the-ai-mind-anthropic-researchers-peer-inside-the-black-box\\\/\",\"url\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/05\\\/decoding-the-ai-mind-anthropic-researchers-peer-inside-the-black-box\\\/\",\"name\":\"Decoding the AI mind: Anthropic researchers peer inside the \\\"black box\\\" | DailyAI\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/05\\\/decoding-the-ai-mind-anthropic-researchers-peer-inside-the-black-box\\\/#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/05\\\/decoding-the-ai-mind-anthropic-researchers-peer-inside-the-black-box\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/dailyai.com\\\/wp-content\\\/uploads\\\/2024\\\/05\\\/DALL\u00b7E-2024-05-22-19.15.50-Anthropic-researchers-studying-artificial-intelligence-inside-a-high-tech-laboratory-with-a-prominent-black-box-in-the-center.-The-scene-features-adva.webp\",\"datePublished\":\"2024-05-22T18:16:07+00:00\",\"dateModified\":\"2024-05-24T10:16:44+00:00\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/05\\\/decoding-the-ai-mind-anthropic-researchers-peer-inside-the-black-box\\\/#breadcrumb\"},\"inLanguage\":\"sv-SE\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/dailyai.com\\\/2024\\\/05\\\/decoding-the-ai-mind-anthropic-researchers-peer-inside-the-black-box\\\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"sv-SE\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/05\\\/decoding-the-ai-mind-anthropic-researchers-peer-inside-the-black-box\\\/#primaryimage\",\"url\":\"https:\\\/\\\/dailyai.com\\\/wp-content\\\/uploads\\\/2024\\\/05\\\/DALL\u00b7E-2024-05-22-19.15.50-Anthropic-researchers-studying-artificial-intelligence-inside-a-high-tech-laboratory-with-a-prominent-black-box-in-the-center.-The-scene-features-adva.webp\",\"contentUrl\":\"https:\\\/\\\/dailyai.com\\\/wp-content\\\/uploads\\\/2024\\\/05\\\/DALL\u00b7E-2024-05-22-19.15.50-Anthropic-researchers-studying-artificial-intelligence-inside-a-high-tech-laboratory-with-a-prominent-black-box-in-the-center.-The-scene-features-adva.webp\",\"width\":1792,\"height\":1024},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/2024\\\/05\\\/decoding-the-ai-mind-anthropic-researchers-peer-inside-the-black-box\\\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\\\/\\\/dailyai.com\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Decoding the AI mind: Anthropic researchers peer inside the &#8220;black box&#8221;\"}]},{\"@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\":\"sv-SE\"},{\"@type\":\"Organization\",\"@id\":\"https:\\\/\\\/dailyai.com\\\/#organization\",\"name\":\"DailyAI\",\"url\":\"https:\\\/\\\/dailyai.com\\\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"sv-SE\",\"@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\":\"sv-SE\",\"@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\\\/sv\\\/author\\\/samjeans\\\/\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"Avkodning av AI-sinnet: Antropiska forskare tittar in i den \"svarta l\u00e5dan\" | DailyAI","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\/sv\/2024\/05\/decoding-the-ai-mind-anthropic-researchers-peer-inside-the-black-box\/","og_locale":"sv_SE","og_type":"article","og_title":"Decoding the AI mind: Anthropic researchers peer inside the \"black box\" | DailyAI","og_description":"Anthropic researchers successfully identified millions of concepts within Claude Sonnet, one of their advanced LLMs. AI models are often considered black boxes, meaning you can&#8217;t &#8216;see&#8217; inside them to understand exactly how they work.\u00a0 When you provide an LLM with an input, it generates a response, but the reasoning behind its choices isn\u2019t clear. Your input goes in, and the output comes out \u2013 and even the AI developers themselves don&#8217;t truly understand what happens inside that &#8216;box.&#8217;\u00a0 Neural networks create their own internal representations of information when they map inputs to outputs during data training.\u00a0The building blocks of this","og_url":"https:\/\/dailyai.com\/sv\/2024\/05\/decoding-the-ai-mind-anthropic-researchers-peer-inside-the-black-box\/","og_site_name":"DailyAI","article_published_time":"2024-05-22T18:16:07+00:00","article_modified_time":"2024-05-24T10:16:44+00:00","og_image":[{"width":1792,"height":1024,"url":"https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/DALL\u00b7E-2024-05-22-19.15.50-Anthropic-researchers-studying-artificial-intelligence-inside-a-high-tech-laboratory-with-a-prominent-black-box-in-the-center.-The-scene-features-adva.webp","type":"image\/webp"}],"author":"Sam Jeans","twitter_card":"summary_large_image","twitter_creator":"@DailyAIOfficial","twitter_site":"@DailyAIOfficial","twitter_misc":{"Skriven av":"Sam Jeans","Ber\u00e4knad l\u00e4stid":"6 minuter"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"NewsArticle","@id":"https:\/\/dailyai.com\/2024\/05\/decoding-the-ai-mind-anthropic-researchers-peer-inside-the-black-box\/#article","isPartOf":{"@id":"https:\/\/dailyai.com\/2024\/05\/decoding-the-ai-mind-anthropic-researchers-peer-inside-the-black-box\/"},"author":{"name":"Sam Jeans","@id":"https:\/\/dailyai.com\/#\/schema\/person\/711e81f945549438e8bbc579efdeb3c9"},"headline":"Decoding the AI mind: Anthropic researchers peer inside the &#8220;black box&#8221;","datePublished":"2024-05-22T18:16:07+00:00","dateModified":"2024-05-24T10:16:44+00:00","mainEntityOfPage":{"@id":"https:\/\/dailyai.com\/2024\/05\/decoding-the-ai-mind-anthropic-researchers-peer-inside-the-black-box\/"},"wordCount":1075,"publisher":{"@id":"https:\/\/dailyai.com\/#organization"},"image":{"@id":"https:\/\/dailyai.com\/2024\/05\/decoding-the-ai-mind-anthropic-researchers-peer-inside-the-black-box\/#primaryimage"},"thumbnailUrl":"https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/DALL\u00b7E-2024-05-22-19.15.50-Anthropic-researchers-studying-artificial-intelligence-inside-a-high-tech-laboratory-with-a-prominent-black-box-in-the-center.-The-scene-features-adva.webp","keywords":["Anthropic","machine learning"],"articleSection":["Industry"],"inLanguage":"sv-SE"},{"@type":"WebPage","@id":"https:\/\/dailyai.com\/2024\/05\/decoding-the-ai-mind-anthropic-researchers-peer-inside-the-black-box\/","url":"https:\/\/dailyai.com\/2024\/05\/decoding-the-ai-mind-anthropic-researchers-peer-inside-the-black-box\/","name":"Avkodning av AI-sinnet: Antropiska forskare tittar in i den \"svarta l\u00e5dan\" | DailyAI","isPartOf":{"@id":"https:\/\/dailyai.com\/#website"},"primaryImageOfPage":{"@id":"https:\/\/dailyai.com\/2024\/05\/decoding-the-ai-mind-anthropic-researchers-peer-inside-the-black-box\/#primaryimage"},"image":{"@id":"https:\/\/dailyai.com\/2024\/05\/decoding-the-ai-mind-anthropic-researchers-peer-inside-the-black-box\/#primaryimage"},"thumbnailUrl":"https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/DALL\u00b7E-2024-05-22-19.15.50-Anthropic-researchers-studying-artificial-intelligence-inside-a-high-tech-laboratory-with-a-prominent-black-box-in-the-center.-The-scene-features-adva.webp","datePublished":"2024-05-22T18:16:07+00:00","dateModified":"2024-05-24T10:16:44+00:00","breadcrumb":{"@id":"https:\/\/dailyai.com\/2024\/05\/decoding-the-ai-mind-anthropic-researchers-peer-inside-the-black-box\/#breadcrumb"},"inLanguage":"sv-SE","potentialAction":[{"@type":"ReadAction","target":["https:\/\/dailyai.com\/2024\/05\/decoding-the-ai-mind-anthropic-researchers-peer-inside-the-black-box\/"]}]},{"@type":"ImageObject","inLanguage":"sv-SE","@id":"https:\/\/dailyai.com\/2024\/05\/decoding-the-ai-mind-anthropic-researchers-peer-inside-the-black-box\/#primaryimage","url":"https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/DALL\u00b7E-2024-05-22-19.15.50-Anthropic-researchers-studying-artificial-intelligence-inside-a-high-tech-laboratory-with-a-prominent-black-box-in-the-center.-The-scene-features-adva.webp","contentUrl":"https:\/\/dailyai.com\/wp-content\/uploads\/2024\/05\/DALL\u00b7E-2024-05-22-19.15.50-Anthropic-researchers-studying-artificial-intelligence-inside-a-high-tech-laboratory-with-a-prominent-black-box-in-the-center.-The-scene-features-adva.webp","width":1792,"height":1024},{"@type":"BreadcrumbList","@id":"https:\/\/dailyai.com\/2024\/05\/decoding-the-ai-mind-anthropic-researchers-peer-inside-the-black-box\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/dailyai.com\/"},{"@type":"ListItem","position":2,"name":"Decoding the AI mind: Anthropic researchers peer inside the &#8220;black box&#8221;"}]},{"@type":"WebSite","@id":"https:\/\/dailyai.com\/#website","url":"https:\/\/dailyai.com\/","name":"DagligaAI","description":"Din dagliga dos av AI-nyheter","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":"sv-SE"},{"@type":"Organization","@id":"https:\/\/dailyai.com\/#organization","name":"DagligaAI","url":"https:\/\/dailyai.com\/","logo":{"@type":"ImageObject","inLanguage":"sv-SE","@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":"sv-SE","@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 \u00e4r en vetenskaps- och teknikskribent som har arbetat i olika AI-startups. N\u00e4r han inte skriver l\u00e4ser han medicinska tidskrifter eller gr\u00e4ver igenom l\u00e5dor med vinylskivor.","sameAs":["https:\/\/www.linkedin.com\/in\/sam-jeans-6746b9142\/"],"url":"https:\/\/dailyai.com\/sv\/author\/samjeans\/"}]}},"_links":{"self":[{"href":"https:\/\/dailyai.com\/sv\/wp-json\/wp\/v2\/posts\/12519","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/dailyai.com\/sv\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/dailyai.com\/sv\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/dailyai.com\/sv\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/dailyai.com\/sv\/wp-json\/wp\/v2\/comments?post=12519"}],"version-history":[{"count":11,"href":"https:\/\/dailyai.com\/sv\/wp-json\/wp\/v2\/posts\/12519\/revisions"}],"predecessor-version":[{"id":12573,"href":"https:\/\/dailyai.com\/sv\/wp-json\/wp\/v2\/posts\/12519\/revisions\/12573"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/dailyai.com\/sv\/wp-json\/wp\/v2\/media\/12522"}],"wp:attachment":[{"href":"https:\/\/dailyai.com\/sv\/wp-json\/wp\/v2\/media?parent=12519"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/dailyai.com\/sv\/wp-json\/wp\/v2\/categories?post=12519"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/dailyai.com\/sv\/wp-json\/wp\/v2\/tags?post=12519"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}