[{"data":1,"prerenderedAt":398},["ShallowReactive",2],{"project-47886517":3},{"id":4,"name":5,"line":6,"line_source":7,"title":8,"url":9,"domain":10,"maker":11,"posted_at":12,"year":13,"points":14,"comments":15,"rank_in_year":16,"shown_in_year":17,"top":18,"tier":19,"category":20,"subcategory":21,"format":22,"keywords":23,"reception":28,"sentiment":29,"stance_mix":30,"scored_comments":35,"aspects":36,"strip":43,"post_text":44,"what":45,"said":65,"maker_in_thread":199,"shape":208,"grid":212,"attempts":360,"same_week":361,"maker_shows":33,"maker_hidden":390,"reader_version":391,"card":392},"47886517","How LLMs Work","Interactive visual guide based on Karpathy's lecture","title","Show HN: How LLMs Work – Interactive visual guide based on Karpathy's lecture","https:\u002F\u002Fynarwal.github.io\u002Fhow-llms-work\u002F","ynarwal.github.io","ynarwal__",1777013333,2026,245,55,199,4366,5,"notable","learning_science","interactive explainer","web_app",[24,25,26,27],"llm explainer","karpathy lecture","how llms work","transformer visualization","roasted",-0.527,{"positive":31,"neutral":32,"mixed":33,"critical":34},3,4,1,22,30,{"idea":37,"execution":39,"design":40,"docs":42},{"positive":38,"critical":31},0,{"positive":33,"critical":18},{"positive":38,"critical":41},9,{"positive":33,"critical":31},[33,33,33,33,33,33,38,38,38,33,38,38,38,38,38,38,31,38,38,38,38,33,33,33,33,33,38,33,38,33,31,33,18,33,31,33,33,33,38,31],"All content is based on Andrej Karpathy's \"Intro to Large Language Models\" lecture (youtube.com\u002Fwatch?v=7xTGNNLPyMI). I downloaded the transcript and used Claude Code to generate the entire interactive site from it — single HTML file. I find it useful to revisit this content time to time.",{"problem":46,"audience":52,"pricing":52,"motivation":53,"time_to_build":52,"stack":57,"openness":62,"stage":63,"team_size":64},{"value":47,"source":48,"evidence":49},"Explaining how large language models work in an interactive visual format","maker",[50],{"id":51},"post",null,{"value":54,"source":48,"evidence":55},"Revisiting Karpathy's lecture content from time to time",[56],{"id":51},[58],{"value":59,"source":48,"evidence":60},"Claude Code generated single HTML file",[61],{"id":51},"unknown","launched","solo",{"praise":66,"objections":86,"requests":146,"questions":165,"alternatives":177},[67,75],{"value":68,"source":69,"evidence":70},"Proper attribution to Karpathy's lecture","commenter",[71],{"id":72,"by":73,"text":74},"47889922","gslepak","Just want to give appreciation for proper attribution. I feel like still some people will say \"Here's something I made\" when the reality is, \"Here's something I asked my AI to…",{"value":76,"source":69,"evidence":77},"Well explained and useful",[78,82],{"id":79,"by":80,"text":81},"47886602","learningToFly33","I’ve had a look, and it’s very well explained! If you ever want to expand it, you could also add how embedded data is fed at the very final step for specific tasks, and how it…",{"id":83,"by":84,"text":85},"47891445","gn_central","This is really well done. Curious how long it took to build something this interactive?",[87,106,114,124,135],{"value":88,"source":69,"evidence":89},"Content appears entirely AI-generated with little review",[90,94,98,102],{"id":91,"by":92,"text":93},"47887999","Udo","> What does the input side of the neutral network look like? Is it enough bits to represent N tokens where N is the context size? Not quite. The raw text converted into IDs…",{"id":95,"by":96,"text":97},"47888036","PetitPrince","Have you reread what was produced by Claude Code before publishing ? This thing in one of the first paragraph jumps out: > you end up with about 44 terabytes — roughly what fits…",{"id":99,"by":100,"text":101},"47888602","arcza","Another low effort, dark mode slopsite. You lost me at \"44 terabytes\" before I even got to the emdash in that sentence. @dang, when is the 'flag as slop' button coming?",{"id":103,"by":104,"text":105},"47892405","jblitzar","yeah this seems entirely LLM-generated.",{"value":107,"source":69,"evidence":108},"Factual error about 44 TB fitting on a single hard drive",[109,110],{"id":95,"by":96,"text":97},{"id":111,"by":112,"text":113},"47889414","embedding-shape","Even if those things are true, no reasonable person who knows what they're talking about, and who writes for an wider audience would say something like \"you end up with about 44…",{"value":115,"source":69,"evidence":116},"Generic AI-style CSS design",[117,121],{"id":118,"by":119,"text":120},"47888607","endymion-light","I really dislike the default AI slop css - if you're going to do this - please have a design language and taste ideas beforehand. It can help so much in refining the look.…",{"id":122,"by":104,"text":123},"47892427","I think the css here is what Claude spits out when you prompt it to make frontend",{"value":125,"source":69,"evidence":126},"Tokenization example is misleading",[127,131],{"id":128,"by":129,"text":130},"47891050","thesz","The page does very poor job tokenizing phrase \"Noinceolik fiyulnabmed fyvaproldge\" into \"Noinceolik fiyulnabm ed fyvaproldge\", factoring only \"ed\" suffix. As if made up words…",{"id":132,"by":133,"text":134},"47891193","vova_hn2","I think that BPE visualization is slightly misleading, because it seems to imply that the \"old\" (smaller) tokens are thrown away and replaced with longer tokens, which is not the…",{"value":136,"source":69,"evidence":137},"Scroll jumping and layout overlap on iOS and narrow windows",[138,142],{"id":139,"by":140,"text":141},"47887415","lukeholder","Page keeps annoyingly scroll-jumping a few pixels on iOS safari",{"id":143,"by":144,"text":145},"47887782","Barbing","Lefthand labels (like Introduction) can overlap over main text content on the right in the central panel - may be able to trigger by reducing window width.",[147,154,158],{"value":148,"source":69,"evidence":149},"Explain embeddings and input handling more thoroughly",[150],{"id":151,"by":152,"text":153},"47887561","gushogg-blake","I haven't found an explanation yet that answers a couple of seemingly basic questions about the LLMs: What does the input side of the neutral network look like? Is it enough bits…",{"value":155,"source":69,"evidence":156},"Cover how embeddings feed into final prediction tasks",[157],{"id":79,"by":80,"text":81},{"value":159,"source":69,"evidence":160},"Move tokenization to the end and start with byte-level model",[161],{"id":162,"by":163,"text":164},"47892648","amelius","In these types of tutorial, can't we move tokenization to the end and just start with a model that inputs and outputs bytes (latin-1) instead? That would (1) take the reader to…",[166,173],{"value":167,"source":69,"evidence":168},"Was this created with Claude Design?",[169],{"id":170,"by":171,"text":172},"47891209","dylkil","Was this created with claude design? it looks very similar to something it mocked for me last week",{"value":174,"source":69,"evidence":175},"How long did it take to build?",[176],{"id":83,"by":84,"text":85},[178,185,192],{"value":179,"source":69,"evidence":180},"Jay Alammar's The Illustrated GPT-2",[181],{"id":182,"by":183,"text":184},"47891430","jasonjmcghee","Highly recommend instead reading the human created \"The Illustrated GPT-2\" by Jay Alammar - https:\u002F\u002Fjalammar.github.io\u002Fillustrated-gpt2\u002F And his similar work. He also has a free…",{"value":186,"source":69,"evidence":187},"Karpathy's original lecture",[188],{"id":189,"by":190,"text":191},"47888195","QuantumNomad_","I agreed when I read your comment, but that turned out to be almost directly from the video https:\u002F\u002Fyoutu.be\u002F7xTGNNLPyMI From around the 2:20 mark he says: “[…] actually ends up…",{"value":193,"source":69,"evidence":194},"tiktokenizer \u002F gpt-tokenizer.dev",[195,196],{"id":128,"by":129,"text":130},{"id":197,"by":11,"text":198},"47891139","I appreciate the feedback, I did notice that as well and I had this thought perhaps this is not worth fixing since I have a link to tiktokenizer. I decided to remove it and just…",{"comments":31,"answered":38,"key":200},[201,204,205],{"id":202,"text":203},"47890941","Update: The \"single hard drive\" claim was wrong and I've corrected it to \"roughly 10 consumer hard drives\" (44TB ÷ ~4TB = ~11). Attribution to Karpathy is now a direct link.…",{"id":197,"text":198},{"id":206,"text":207},"47891239","I disagree with some comments saying it's not worth reading since it's generated by LLM. Even though I made it clear that I have download the transcript. LLMs are exceptionally…",{"commenters":209,"max_depth":210,"maker_replies":31,"first_comment_mins":41,"lifespan_hours":211},41,8,28,[213,217,221,223,226,228,231,235,236,238,241,244,247,250,253,255,258,260,262,264,266,269,271,274,276,277,280,282,284,287,289,291,294,297,300,303,306,307,309,311,312,314,316,319,321,325,328,331,333,337,339,341,342,346,350,354,358],[214,38,33,215,216,38],"47888179","project\u002Fcritical\u002Fnone","lateral_cloud",[218,33,33,219,220,38],"47892297","project\u002Fcritical\u002Fidea","skiing_crawling",[103,33,33,222,104,38],"project\u002Fcritical\u002Fexecution",[170,224,31,225,171,38],2,"project\u002Fneutral\u002Fdesign",[99,31,33,227,100,38],"project\u002Fcritical\u002Fdesign",[229,31,33,215,230,38],"47888651","rrr_oh_man",[232,31,38,233,234,38],"47889561","other\u002Fneutral\u002Fnone","rahen",[95,32,33,222,96,38],[189,32,38,237,190,38],"commenter\u002Fneutral\u002Fexecution",[239,32,38,233,240,38],"47888316","erwincoumans",[242,32,38,233,243,38],"47890316","sigmoid10",[245,32,38,233,246,38],"47890965","littlestymaar",[248,32,38,233,249,38],"47888938","lucideer",[251,32,33,222,252,38],"47889273","ashtonshears",[111,32,38,254,112,38],"commenter\u002Fcritical\u002Fexecution",[256,32,38,257,252,38],"47889441","commenter\u002Fneutral\u002Fnone",[259,32,38,257,112,38],"47889786",[261,32,38,257,252,38],"47890132",[263,32,38,257,112,38],"47890655",[265,32,38,257,252,38],"47900329",[267,32,38,257,268,38],"47890951","raincole",[270,32,38,257,112,38],"47891282",[272,32,38,237,273,38],"47889555","tcp_handshaker",[151,18,31,275,152,38],"project\u002Fneutral\u002Fdocs",[91,18,38,233,92,38],[278,18,38,233,279,38],"47888127","maciejzj",[281,18,38,233,152,38],"47889844",[283,18,38,233,243,38],"47890418",[285,18,38,233,286,38],"47888201","GistNoesis",[288,18,38,233,152,38],"47889877",[139,290,33,227,140,38],6,[292,290,33,227,293,38],"47887823","tbreschi",[295,290,33,227,296,38],"47890214","vinnymac",[298,290,31,225,299,38],"47890280","skinner927",[182,301,33,302,183,38],7,"project\u002Fcritical\u002Fdocs",[304,301,38,305,104,38],"47892410","other\u002Fpositive\u002Fnone",[132,210,33,222,133,38],[202,41,38,308,11,33],"maker\u002Fneutral\u002Fexecution",[128,310,33,222,129,38],10,[197,310,38,308,11,33],[313,310,38,233,129,38],"47900177",[143,315,33,227,144,38],11,[206,317,38,318,11,33],12,"maker\u002Fneutral\u002Fnone",[162,320,31,275,163,38],13,[322,323,33,302,324,38],"47891440",14,"siva7",[72,326,18,327,73,38],15,"maker\u002Fpositive\u002Fnone",[79,329,18,330,80,38],16,"project\u002Fpositive\u002Fdocs",[332,329,33,302,230,38],"47888604",[334,329,31,335,336,38],"47890631","project\u002Fmixed\u002Fnone","davkap92",[118,338,33,227,119,38],17,[340,338,33,227,230,38],"47888656",[122,338,33,227,104,38],[343,344,33,219,345,38],"47888674",18,"5asHajh",[347,348,33,227,349,38],"47890345",19,"weego",[351,352,38,233,353,38],"47888368",20,"PeakScripter",[355,356,33,219,357,38],"47888740",21,"hansmayer",[83,34,18,359,84,38],"project\u002Fpositive\u002Fexecution",[],[362,367,371,376,380,385],{"id":363,"name":364,"points":365,"category":366},"47920787","dirac",393,"dev_tools",{"id":368,"name":369,"points":370,"category":366},"47874647","Honker",327,{"id":372,"name":373,"points":374,"category":375},"47882697","Tolaria",318,"productivity",{"id":377,"name":378,"points":379,"category":375},"47899844","WUPHF",260,{"id":381,"name":382,"points":383,"category":384},"47937380","Auto-Architecture",241,"hardware",{"id":386,"name":387,"points":388,"category":389},"47876071","Turning a Gaussian Splat into a videogame",236,"games",false,"v2",{"id":393,"name":5,"line":6,"ai":390,"cat":20,"year":13,"points":14,"comments":15,"rec":28,"s":394,"top":18,"kw":395,"maker":11,"domain":10,"tier":19,"attempts":38,"strip":396,"url":9,"mix":397,"praise":68,"objection":88,"sub":21},47886517,-0.53,[24,25,26,27],"1111110001000000300001111101013151311103",[31,32,33,34],1794036699361]