{"success":true,"count":2,"items":[{"videoId":"iXd0t60YmMw","chunkIndex":0,"totalChunks":2,"title":"Karpathy's LLM Wiki - Full Beginner Setup Guide — Part 1 of 2","thumbnail":"https://i.ytimg.com/vi/iXd0t60YmMw/maxresdefault.jpg","duration":905,"uploader":"Teacher's Tech","youtubeUrl":"https://www.youtube.com/watch?v=iXd0t60YmMw","keywords":["ai","rag","knowledge-base","obsidian","markdown","llm","productivity","note-taking","automation","knowledge-management"],"normalizedKeywords":["엔지니어링","프로덕트","기술 트렌드"],"targetAudience":[{"who":"지식노동자","why":"문서와 정보를 자주 다루는 사람에게 AI 지식관리 방식이 유용함"},{"who":"개발 입문자","why":"비전공자도 따라할 수 있는 파일 기반 워크플로를 보여줌"},{"who":"콘텐츠 정리자","why":"여러 자료를 구조화해 재사용하는 방법을 배울 수 있음"}],"normalizedAudience":["지식노동자 일반","학생·주니어","엔지니어·개발자"],"summary":"이 영상은 기존 RAG 방식의 한계, 즉 질문할 때마다 문서를 매번 다시 검색하고 조합해야 해서 지식이 축적되지 않는 문제를 짚고, 이를 해결하는 Andre Karpathy의 'LLM wiki' 개념을 소개한다. 핵심은 AI가 원본 문서를 한 번 읽고, 그 내용을 서로 연결된 마크다운 위키로 정리해 누적되는 지식베이스를 만드는 것이다.\n\n영상은 Obsidian과 Claude Code 같은 도구로 이를 실제로 구현하는 과정을 보여준다. raw/source, wiki, schema라는 3층 구조를 만들고, 규칙 파일을 통해 AI가 새 문서를 읽어 위키를 업데이트하고, 요약·링크·충돌 감지까지 수행하게 한다. 결과적으로 사용자는 매번 처음부터 검색하는 대신, 이미 정리된 지식망 위에서 질문하고 탐색할 수 있게 된다.","insights":["RAG의 한계는 검색이 아니라 '축적 부재'다.","지식은 읽는 순간보다 연결될 때 더 큰 가치를 만든다.","원본과 파생 위키를 분리해야 지식베이스가 안정적이다.","규칙(schema)이 있어야 AI가 일관된 편집자로 동작한다.","문서 관리의 미래는 저장보다 구조화와 업데이트다."],"keyClips":[{"clipId":"iXd0t60YmMw:c0:1-27","startSegmentIndex":1,"endSegmentIndex":27,"startTime":0,"endTime":160.70125,"durationSeconds":160.7,"preview":"RAG의 한계와 위키","mustSee":true},{"clipId":"iXd0t60YmMw:c0:28-38","startSegmentIndex":28,"endSegmentIndex":38,"startTime":160.70000000000002,"endTime":229.99499999999998,"durationSeconds":69.3,"preview":"LLM 역할 분담","mustSee":false},{"clipId":"iXd0t60YmMw:c0:39-55","startSegmentIndex":39,"endSegmentIndex":55,"startTime":229.74,"endTime":350.04366666666664,"durationSeconds":120.3,"preview":"폴더 구조 세팅","mustSee":false},{"clipId":"iXd0t60YmMw:c0:56-76","startSegmentIndex":56,"endSegmentIndex":76,"startTime":350.06,"endTime":492.76944444444445,"durationSeconds":142.7,"preview":"규칙 파일의 힘","mustSee":true},{"clipId":"iXd0t60YmMw:c0:77-92","startSegmentIndex":77,"endSegmentIndex":92,"startTime":492.62,"endTime":604.1206666666667,"durationSeconds":111.5,"preview":"첫 문서 적재","mustSee":true}],"curatedSegments":[{"segmentIndex":18,"text":"like it's never read any of those papers before. It starts fresh every time. That's the bottleneck.","startTime":105.7,"endTime":112.02088235294117,"durationSeconds":6,"level":"B2","overallScore":7.6,"rationale":"문제 본질을 날카롭게 요약함."},{"segmentIndex":36,"text":"want more infrastructure than just some markdown files. Second, garbage in, garbage out. The wiki","startTime":834.66,"endTime":840.644,"durationSeconds":6,"level":"B2","overallScore":8,"rationale":"핵심 원칙과 표현 가치가 높음."},{"segmentIndex":44,"text":"And that's the LLM Wiki, a personal knowledge base that the AI builds and maintains for you that actually gets better over time instead of starting from scratch on every question.","startTime":879.86,"endTime":889.8892857142857,"durationSeconds":10,"level":"B2","overallScore":7.8,"rationale":"핵심 가치가 응축된 결론 문장."},{"segmentIndex":5,"text":"all of that work again from scratch. Nothing was saved. Nothing was built up. Every single","startTime":21.18,"endTime":26.102500000000003,"durationSeconds":5,"level":"B2","overallScore":6.8,"rationale":"누적 부재를 강하게 요약함."},{"segmentIndex":16,"text":"accumulation. Nothing compounds. Think about it like this. Imagine you're a researcher and you've","startTime":94.74000000000001,"endTime":100.25857142857143,"durationSeconds":6,"level":"B2","overallScore":7,"rationale":"비유 도입으로 이해를 돕는 문장."},{"segmentIndex":19,"text":"Garpathy's idea flips this completely. Instead of searching raw documents every time you ask a","startTime":111.62,"endTime":116.71000000000001,"durationSeconds":5,"level":"B2","overallScore":6.6,"rationale":"해결책의 관점 전환이 뚜렷함."},{"segmentIndex":27,"text":"the AI is not starting from scratch. It's actually working from a pre-built organized knowledge base.","startTime":154.94,"endTime":160.70125,"durationSeconds":6,"level":"B2","overallScore":7,"rationale":"차별점이 명확한 핵심 문장."},{"segmentIndex":26,"text":"So, what would you actually use this for? Here are some ideas. If you're a student or a researcher, build a wiki as you read papers and articles on a topic. By the end, you have this structured","startTime":772.74,"endTime":782.8110526315791,"durationSeconds":10,"level":"B2","overallScore":7,"rationale":"활용처와 결과를 잘 제시함."},{"segmentIndex":42,"text":"all that said, this is genuinely one of the most practical AI workflows I've seen. It solves real","startTime":867.54,"endTime":873.09,"durationSeconds":6,"level":"B2","overallScore":6.8,"rationale":"평가와 핵심 장점을 강하게 제시함."},{"segmentIndex":10,"text":"why it matters, and then we're going to build one together from scratch, step by step. You don't need to be technical. If you can create a folder on your computer, you can do this.","startTime":48.339999999999996,"endTime":57.727777777777774,"durationSeconds":9,"level":"A2","overallScore":4.4,"rationale":""},{"segmentIndex":15,"text":"to find all those pieces and stitch them together every single time. There's no memory. There's no","startTime":88.78,"endTime":94.79,"durationSeconds":6,"level":"B1","overallScore":4.8,"rationale":""},{"segmentIndex":20,"text":"question, you have the AI read your documents once and build a structured wiki out of them.","startTime":116.66000000000001,"endTime":122.24852941176472,"durationSeconds":6,"level":"B2","overallScore":5,"rationale":""},{"segmentIndex":28,"text":"Here's how Kaproy describes it. He says, \"Think of Obsidian as the IDE, the LLM as the programmer,","startTime":160.70000000000002,"endTime":167.8361111111111,"durationSeconds":7,"level":"C1","overallScore":5,"rationale":""},{"segmentIndex":74,"text":"I'm giving you gives you a solid starting point. You can always refine it as you go. That's actually part of the process. The schema evolves as the wiki grows. I'm also going to add","startTime":468.14,"endTime":478.1894444444445,"durationSeconds":10,"level":"B1","overallScore":4.6,"rationale":""},{"segmentIndex":33,"text":"The pattern works anywhere you're accumulating knowledge over time and you want it organized rather than just scattered. Okay, let me be straight about the limitations because this","startTime":814.38,"endTime":823.8371428571429,"durationSeconds":9,"level":"B2","overallScore":4.6,"rationale":""},{"segmentIndex":1,"text":"you can see all the different wiki pages here that it's planning to create. And if this looks good, I'm going to tell it to go ahead. But you can see how I can adjust the scope as well.","startTime":603.66,"endTime":612.839,"durationSeconds":9,"level":"B1","overallScore":4,"rationale":""},{"segmentIndex":16,"text":"the week from the neighborhood pages, the food pages, the temple page. It's connecting dots that were spread across completely different sources. It's citing specific wiki pages. This","startTime":703.22,"endTime":712.9523076923077,"durationSeconds":10,"level":"B2","overallScore":3.8,"rationale":""}],"generatedAt":"2026-06-24T23:45:17.873Z","keyClipsTotalSec":620},{"videoId":"iXd0t60YmMw","chunkIndex":1,"totalChunks":2,"title":"Karpathy's LLM Wiki - Full Beginner Setup Guide — Part 2 of 2","thumbnail":"https://i.ytimg.com/vi/iXd0t60YmMw/maxresdefault.jpg","duration":905,"uploader":"Teacher's Tech","youtubeUrl":"https://www.youtube.com/watch?v=iXd0t60YmMw","keywords":["llm","knowledge-base","obsidian","ai-workflow","note-taking","rag","information-management","personal-productivity"],"normalizedKeywords":["교육","엔지니어링","커리어·성장"],"targetAudience":[{"who":"지식노동자","why":"흩어진 자료를 구조화해 개인 지식베이스로 만들고 싶은 사람에게 유용함"},{"who":"학생·연구자","why":"읽는 자료를 누적하며 연결된 형태로 정리하는 방법을 배울 수 있음"},{"who":"팀 문서 담당자","why":"회의록·프로젝트 문서를 검색 가능한 위키로 바꾸는 아이디어를 얻을 수 있음"}],"normalizedAudience":["지식노동자 일반","학생·주니어","리서처·학자"],"summary":"이 영상은 Claude Code와 Obsidian을 활용해 개인용 LLM 위키를 만드는 방법을 보여준다. 단순히 문서를 저장하는 수준이 아니라, AI가 여러 소스를 읽고 구조화된 페이지를 만들며, 새 자료가 추가되면 기존 페이지까지 갱신해 지식 간 연결을 점점 더 촘촘하게 만든다는 점이 핵심이다.\n\n또한 서로 다른 출처의 정보를 조합해 질문에 답하게 하고, 위키 전체를 점검하는 'lint' 개념으로 모순·고아 페이지·깨진 링크를 찾아내는 유지보수 방식도 소개한다. 마지막에는 학생, 교사, 비즈니스 팀, 독서가 많은 개인 등 다양한 활용 사례와 함께, 개인 규모에 적합하고 좋은 자료를 넣어야 하며 AI 오류를 검토해야 한다는 한계도 짚는다.","insights":["핵심 가치는 저장이 아니라 연결이다.","새 자료가 들어오면 기존 지식도 함께 갱신돼야 한다.","AI 위키는 검색보다 추론형 질의에 더 강하다.","lint는 지식베이스의 품질 관리 장치다.","개인 규모에선 plain text 위키가 가장 실용적이다."],"keyClips":[{"clipId":"iXd0t60YmMw:c1:1-6","startSegmentIndex":1,"endSegmentIndex":6,"startTime":603.66,"endTime":642.8186666666667,"durationSeconds":39.2,"preview":"구조화된 위키 생성","mustSee":false},{"clipId":"iXd0t60YmMw:c1:7-17","startSegmentIndex":7,"endSegmentIndex":17,"startTime":642.54,"endTime":717.8266666666667,"durationSeconds":75.3,"preview":"새 자료가 들어올 때","mustSee":true},{"clipId":"iXd0t60YmMw:c1:18-25","startSegmentIndex":18,"endSegmentIndex":25,"startTime":717.66,"endTime":772.6905555555556,"durationSeconds":55,"preview":"위키 lint의 역할","mustSee":true},{"clipId":"iXd0t60YmMw:c1:26-32","startSegmentIndex":26,"endSegmentIndex":32,"startTime":772.74,"endTime":814.8236666666667,"durationSeconds":42.1,"preview":"실전 활용 시나리오","mustSee":false},{"clipId":"iXd0t60YmMw:c1:33-44","startSegmentIndex":33,"endSegmentIndex":44,"startTime":814.38,"endTime":889.8892857142857,"durationSeconds":75.5,"preview":"한계와 현실 조건","mustSee":true}],"curatedSegments":[{"segmentIndex":18,"text":"like it's never read any of those papers before. It starts fresh every time. That's the bottleneck.","startTime":105.7,"endTime":112.02088235294117,"durationSeconds":6,"level":"B2","overallScore":7.6,"rationale":"문제 본질을 날카롭게 요약함."},{"segmentIndex":36,"text":"want more infrastructure than just some markdown files. Second, garbage in, garbage out. The wiki","startTime":834.66,"endTime":840.644,"durationSeconds":6,"level":"B2","overallScore":8,"rationale":"핵심 원칙과 표현 가치가 높음."},{"segmentIndex":44,"text":"And that's the LLM Wiki, a personal knowledge base that the AI builds and maintains for you that actually gets better over time instead of starting from scratch on every question.","startTime":879.86,"endTime":889.8892857142857,"durationSeconds":10,"level":"B2","overallScore":7.8,"rationale":"핵심 가치가 응축된 결론 문장."},{"segmentIndex":5,"text":"all of that work again from scratch. Nothing was saved. Nothing was built up. Every single","startTime":21.18,"endTime":26.102500000000003,"durationSeconds":5,"level":"B2","overallScore":6.8,"rationale":"누적 부재를 강하게 요약함."},{"segmentIndex":16,"text":"accumulation. Nothing compounds. Think about it like this. Imagine you're a researcher and you've","startTime":94.74000000000001,"endTime":100.25857142857143,"durationSeconds":6,"level":"B2","overallScore":7,"rationale":"비유 도입으로 이해를 돕는 문장."},{"segmentIndex":19,"text":"Garpathy's idea flips this completely. Instead of searching raw documents every time you ask a","startTime":111.62,"endTime":116.71000000000001,"durationSeconds":5,"level":"B2","overallScore":6.6,"rationale":"해결책의 관점 전환이 뚜렷함."},{"segmentIndex":27,"text":"the AI is not starting from scratch. It's actually working from a pre-built organized knowledge base.","startTime":154.94,"endTime":160.70125,"durationSeconds":6,"level":"B2","overallScore":7,"rationale":"차별점이 명확한 핵심 문장."},{"segmentIndex":26,"text":"So, what would you actually use this for? Here are some ideas. If you're a student or a researcher, build a wiki as you read papers and articles on a topic. By the end, you have this structured","startTime":772.74,"endTime":782.8110526315791,"durationSeconds":10,"level":"B2","overallScore":7,"rationale":"활용처와 결과를 잘 제시함."},{"segmentIndex":42,"text":"all that said, this is genuinely one of the most practical AI workflows I've seen. It solves real","startTime":867.54,"endTime":873.09,"durationSeconds":6,"level":"B2","overallScore":6.8,"rationale":"평가와 핵심 장점을 강하게 제시함."},{"segmentIndex":10,"text":"why it matters, and then we're going to build one together from scratch, step by step. You don't need to be technical. If you can create a folder on your computer, you can do this.","startTime":48.339999999999996,"endTime":57.727777777777774,"durationSeconds":9,"level":"A2","overallScore":4.4,"rationale":""},{"segmentIndex":15,"text":"to find all those pieces and stitch them together every single time. There's no memory. There's no","startTime":88.78,"endTime":94.79,"durationSeconds":6,"level":"B1","overallScore":4.8,"rationale":""},{"segmentIndex":20,"text":"question, you have the AI read your documents once and build a structured wiki out of them.","startTime":116.66000000000001,"endTime":122.24852941176472,"durationSeconds":6,"level":"B2","overallScore":5,"rationale":""},{"segmentIndex":28,"text":"Here's how Kaproy describes it. He says, \"Think of Obsidian as the IDE, the LLM as the programmer,","startTime":160.70000000000002,"endTime":167.8361111111111,"durationSeconds":7,"level":"C1","overallScore":5,"rationale":""},{"segmentIndex":74,"text":"I'm giving you gives you a solid starting point. You can always refine it as you go. That's actually part of the process. The schema evolves as the wiki grows. I'm also going to add","startTime":468.14,"endTime":478.1894444444445,"durationSeconds":10,"level":"B1","overallScore":4.6,"rationale":""},{"segmentIndex":33,"text":"The pattern works anywhere you're accumulating knowledge over time and you want it organized rather than just scattered. Okay, let me be straight about the limitations because this","startTime":814.38,"endTime":823.8371428571429,"durationSeconds":9,"level":"B2","overallScore":4.6,"rationale":""},{"segmentIndex":1,"text":"you can see all the different wiki pages here that it's planning to create. And if this looks good, I'm going to tell it to go ahead. But you can see how I can adjust the scope as well.","startTime":603.66,"endTime":612.839,"durationSeconds":9,"level":"B1","overallScore":4,"rationale":""},{"segmentIndex":16,"text":"the week from the neighborhood pages, the food pages, the temple page. It's connecting dots that were spread across completely different sources. It's citing specific wiki pages. This","startTime":703.22,"endTime":712.9523076923077,"durationSeconds":10,"level":"B2","overallScore":3.8,"rationale":""}],"generatedAt":"2026-06-24T23:45:43.254Z","keyClipsTotalSec":620}]}