Managing director of startup incubating program, TIPS Town, Seoul, Korea. 2,734+ startups granted ave. $370k, 20 IPO, 74 M&A, 93+ raised $22M+ funds.
Seoul, South Korea
Joined March 2009
- Tweets6.3K
- Following220
- Followers1.4K
- Likes512
Doer Youngil Ahn retweeted
This paragraph by Richard Feynman hits so hard:
“Fall in love with some activity, and do it! Nobody ever figures out what life is all about, and it doesn’t matter. Explore the world. Nearly everything is really interesting if you go into it deeply enough. Work as hard and as much as you want to on the things you like to do the best. Don’t think about what you want to be, but what you want to do. Keep up some kind of a minimum with other things so that society doesn’t stop you from doing anything at all.”
Doer Youngil Ahn retweeted
Weekend hack: HackerTube. I was annoyed with YouTube's recommendations so I had Claude filter down 43M comments over 20 years of HN to 6,454 videos that keep being recommended. *Significantly* better viewing experience, ~2 hours with Opus 5.5:
hackertube.org/
- Excellent recs without any logins (state is kept in local browser session)
- Thumbnail is a representative frame from middle of the video instead of the super duper A/B test hyper-optimized user uploads
- Underlying data from the comment analysis as one zip file, point your Claude at it and ask for your own version
- HackerTube Longs ✊
A bit more on the process:
I use Claude Code (desktop app for Mac), Opus 5.5 on Medium effort as my default. I only talk to Claude and have it do everything, the only "manual work" I helped with in tandem was in the browser (reminiscent of MenuGen): 1) I had to create some kind of API key from Google Cloud UI to query YouTube for video metadata, 2) I booted up a VPS box for serving and 3) I bought the domain name. Once the box was up I gave Claude the IP and told it update, secure, harden, deploy and optimize the code, and it did the whole thing.
Many awesome projects that would have normally taken weeks/months even can now be few-shotted for breakfast. What else are we not thinking of.
Doer Youngil Ahn retweeted
스티브 잡스 아들내미 근황
"내 평생을 걸고 암 조진다" 선언 ㄷㄷ
2011년에 스티브 잡스가 췌장암으로 떠난 건 다들 알 거임.
그 아들 리드 잡스(Reed Jobs)가 지금 뭐 하고 있나 봤더니, VC(벤처캐피털) 차려서 암 정복하겠다고 인생 올인 박음 ㅋ
펀드 이름은 '요세미티(Yosemite)'인데, 일반적인 투자사들이랑 굴러가는 뽄새가 완전 다름
• 딴 거 안 보고 바이오테크 파이의 40% 먹는 '종양학(Oncology)'에만 올인 몰빵함
• 대학 연구실에 "돈 벌어오라" 조건도 안 달고 쌩으로 연구비 꽂아줘서 초기 리스크 다 지워줌
• 예일, 버클리, 스탠포드 석학들이랑 붙어서 아예 회사 3분의 1을 자체적으로 직접 빌딩해버림
• 펀드 1호로 20개 기업 꽂았고, 펀드 2호는 목표치 3.5억 달러(약 4,800억) 가뿐히 넘겨서 오버서브스크라이브 달성함.
파트너로 애브비 $ABBV 랑 암젠 $AMGN 같은 빅파마 거물들도 묻어둠 ㅎㄷㄷ
기자가 "만약 아버지가 살아계셨다면 뭐라고 하셨을 것 같냐"고 물어보니까,
"하던 거 계속해라... 이건 내 평생을 갈아 넣어야 할 롱텀 플레이(Long-term play)다." 라고 답함.
그리고 가장 빡센 점이 뭐냐고 하니까,
본인도 원래 존나 성격 급한 놈인데 암 정복은 원래 평생 걸리는 작업이라는 걸 인정하고 버티는 게 '진짜 규율(Discipline)'이라고 털어놓음.
인내심이라는 건 그냥 타고난 성격이 아니라,
매 순간 이 악물고 내리는 선택이라는 거임.
잡스 유전자가 딴 데로 안 가고 진짜 멋진 곳으로 향했네 🤯
"Not medical advice"
Doer Youngil Ahn retweeted
I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives. More thoughts below.
Doer Youngil Ahn retweeted
6,000 years of world history on a single timeline. Zoom in. You could spend 30 minutes exploring this.
Doer Youngil Ahn retweeted
Neil Movva (@neilmovva) started his career at Nvidia, working on GPUs and kernels, and has an unusually deep understanding of inference, from software to chips to power.
We spend a lot of time on each of those layers, how they connect, and where the important tradeoffs are.
What makes this conversation special is how detailed it is (like a 401-level class), yet Neil makes it remarkably clear and easy to follow.
Today he runs Sail Research, a company building infrastructure for agents to make tokens as cheap as possible.
We discuss:
- Latency versus throughput
- Why there are no bad chips, only bad pricing
- The end of kernel engineering
- Buying chips and power no one else wants
- New chip architectures
- Nvidia lore + his contrarian view of the company
- Open source and the frontier labs
I learned a ton. Enjoy!
TIMESTAMPS
0:00 Intro
0:38 Building a “Token Factory”
4:21 The Future of Background Agents
13:09 Nvidia and the GPU Stack
23:27 Chips, Memory, and Transformers
36:14 The Future of AI Training Data
44:32 Chip Scarcity and Compute Arbitrage
52:44 Reinventing the AI Data Center
59:01 Power and the “Scavenger Strategy”
1:10:10 Open vs. Closed AI
Doer Youngil Ahn retweeted
Mark Steyn explains the stark demographic reality: a 90% population with 1.3 children per couple and a 10% population with 3.5 children will have roughly the same number of grandchildren in only two generations. The transformation is already underway.(2010)
Doer Youngil Ahn retweeted
6 agent patterns for AI engineers:
(explained with usage)
1) prompt chaining
→ split the task into fixed steps, each one checking the last.
→ use when the task decomposes cleanly and accuracy matters more than latency.
2) routing
→ classify the input first, then send it to the model or tool built for it.
→ use when inputs fall into distinct classes that need different handling.
3) parallelization
→ run several calls at once and merge them, either by splitting the work or voting on the same question.
→ use when subtasks are independent, or when one answer deserves several opinions.
4) orchestrator-workers
→ a lead model decides what the subtasks are at runtime, then delegates them.
→ use when you cannot list the steps in advance.
5) evaluator-optimizer
→ one model writes, another grades, the loop repeats until it passes.
→ use when you have clear criteria and iteration measurably helps.
6) autonomous agent
→ no fixed path. it plans, acts, reads feedback from the environment, and decides when it is done.
→ use when the steps are unknowable and you can afford the cost and the blast radius.
the first five are workflows: you wrote the path. only the last one writes its own.
most production systems people call agents are pattern 1, 2, or 5 with good error handling.
this taxonomy is from Anthropic's own writeup on building effective agents.
full breakdown in the article below.
Doer Youngil Ahn retweeted
I ASKED A BILLIONAIRE HOW HE NEVER PANICS DURING CRASHES. HE SENT ME THIS
No explanation, no context, just the framework.
Price falls 5% → Hold
Price falls 15% → Buy 10%
Price falls 25% → Buy 20%
Price falls 40% → Buy 30%
Price falls 60% → Buy 40%
Price rises 5% → Hold
Price rises 20% → Hold
Price rises 25% → Sell 10%
Price rises 40% → Sell 20%
Price rises 50% → Sell 35%
Price rises 70% → Sell 40%
Price rises 100% → Sell 90%
Always keep 10% as a moonbag
Never go all in, never go all out
Took me one read to understand it.
Took me one crash to believe it.
Follow me - the people who do tend not to regret it
Doer Youngil Ahn retweeted
🚨| CHINA ACABÓ CON EL NEGOCIO DE OCR
Un modelo de 3 mil millones de parámetros del tamaño de un cacahuete puede leer un PDF completo de 100 páginas de una sola vez.
Sin dividir páginas.
Sin pérdida de contexto.
Sin factura en la nube.
Conoce Unlimited-OCR 👇
• Lee documentos completos con una ventana de contexto de 32K
• 93% en benchmarks estándar de análisis OCR (+6 sobre la línea base)
• La tasa de error se mantiene por debajo de 0.11 incluso después de 40+ páginas
• Multilingüe de fábrica
• Se ejecuta 100% localmente en tu hardware
• Soporta Transformers, vLLM, SGLang, Docker, Ollama y llama.cpp
Aquí viene lo loco:
La mayoría de las herramientas OCR todavía procesan los documentos página por página.
Unlimited-OCR lee el documento completo como un contexto único, preservando tablas, referencias, diseños y relaciones entre páginas.
Eso lo cambia todo.
Mientras tanto, las empresas están pagando:
• $1.50–$15 por 1.000 páginas
• Enviando PDFs sensibles a proveedores de nube
• Esperando respuestas de API
Este modelo lo hace offline.
Gratis.
Para siempre.
Construido por Baidu para ir más allá de DeepSeek-OCR.
Ya tiene más de 1,9 millones de descargas en Hugging Face...
...y casi nadie está hablando de ello todavía.
El código abierto se mueve más rápido que la mayoría del software empresarial.
Doer Youngil Ahn retweeted
A guy in China plugged an LLM into Karpathy method + Claude Code and built a second brain that remembers everything forever.
Not a note app, not RAG just 3 folders on a laptop.
Folder 1 holds raw sources: articles, journal entries, chat logs. The LLM reads them but never touches them.
Folder 2 is the wiki hundreds of markdown pages the LLM writes and maintains itself, where every person, idea and decision gets filed, cross-referenced and kept current.
Folder 3 is one file: CLAUDEmd, the rulebook that turns a chatbot into a librarian.
He drops in a new article and Claude reads it, updates 10-15 pages, flags what contradicts his old notes, logs the timestamp while he watches his memory grow node by node in Obsidian graph view.
Ask it anything and it doesn’t dig through raw files like ChatGPT does the synthesis already exists, the connections are already drawn.
The method comes from a gist Karpathy posted on April 4, now sitting at 5,000 stars and 5,000 forks, with 1 production team already scaled to 4,000 interlinked concepts in 6 months.
The part nobody mentions: Vannevar Bush described this exact machine in 1945 the Memex, a personal memory store with trails between documents. Nobody built it for 81 years because someone had to do the maintenance.
Now maintenance costs nothing the LLM does the filing while the human does the thinking.
Every article makes the brain smarter, and it never forgets.
His memory has a git history now.
Doer Youngil Ahn retweeted
In order to be born, you needed:
2 parents
4 grandparents
8 great-grandparents
16 second great-grandparents
32 third great-grandparents
64 fourth great-grandparents
128 fifth great-grandparents
256 sixth great-grandparents
512 seventh great-grandparents
1,024 eighth great grandparents
2,048 ninth great-grandparents
For you to be born today from 12 previous generations, you needed a total of 4,094 ancestors over the last 400 years.
Think for a moment:
How many struggles?
How many battles?
How many difficulties?
How much sadness?
How much happiness?
How many love stories?
How many expressions of hope for the future? – did your ancestors have to undergo for you to exist in this present moment...
Doer Youngil Ahn retweeted
Instead of watching another Netflix series tonight, watch this 16-minute talk from a Jane Street AI engineer explaining how the hedge fund uses LLMs to make $30B+ every year.
It’s the clearest, most practical look I’ve seen at how a top quant firm actually builds custom LLMs and agentic tools for developers, especially when public models fall short on something like OCaml.
Whether you’ve never built AI tooling before or you’re already shipping agents, you’ll walk away with concrete ideas on making LLMs reliable in production.
Bookmark it & watch the talk today - it might be the most valuable thing you learn about real-world AI engineering all month.
Doer Youngil Ahn retweeted
Don't pay for ElevenLabs, use Voxilica
Don't pay for Loom Pro, use ScreenDraft
Don't pay for Midjourney, use Glimpse Diffusion
Don't pay for Buffer, use PulseSocial
Don't pay for Notion AI, use ZenBrain
Don't pay for Zapier, use HookFlow
Don't pay for Typeform, use FormForge
Don't pay for Intercom, use ChatSprout
Don't pay for Crunchbase, use VentureMap
Don't pay for Viral Scripting & Video Research, use Syllaby
(SAVE THIS before it disappears)
Doer Youngil Ahn retweeted
El CEO de Anthropic viendo que una Skill gratis de GitHub reduce el uso de tokens de Claude Code en un 90% mientras la gente todavía paga por Max
ESTE REPO HACE QUE CLAUDE ESCRIBA HASTA UN 94% MENOS DE CÓDIGO
Fable 5 es el mejor modelo de código del mundo. También el más caro.
Cada línea que escribe de más, la pagas tú.
Ponytail lo arregla con una idea absurdamente simple: convierte a tu agente en el senior vago del equipo.
Ese que mira tus 50 líneas y las reemplaza por una.
Antes de escribir nada, el agente se pregunta:
→ Esto necesita existir?
→ Ya está en el codebase?
→ Lo hace la librería estándar?
→ Lo hace el navegador de forma nativa?
→ Cabe en una línea?
Solo si todo falla, escribe el mínimo que funciona.
Resultado:
• hasta 94% menos código
• un 20% menos de factura
• un 27% más rápido
Sin recortar seguridad ni validaciones.
Se instala en 2 comandos:
/plugin marketplace add DietrichGebert/ponytail
/plugin install ponytail@ponytail
83.000 estrellas en GitHub.
Gratis y open source.
El mejor código es el que nunca escribiste.
Enlace al repositorio abajo👇
Doer Youngil Ahn retweeted
Un chico en China integró el método de Andrej Karpathy en Claude Code y convirtió un baúl de Obsidian muerto con 15.000 notas en el segundo cerebro más inteligente del planeta.
Su baúl era un cementerio: 956 archivos que nadie volvía a abrir, 80 pestañas guardadas, 0 clics en 6 meses.
Entonces una frase de Karpathy le hizo reaccionar: no sabes nada hasta que puedes construirlo desde cero.
Los tutoriales son un aprendizaje falso que tu cerebro borra en 3 días, junto con las 40 pestañas que guardaste y nunca abriste.
Así que le dio la vuelta a todo el planteamiento.
Obsidian es el IDE, Claude Code es el programador y sus 5.000 notas son el código fuente.
Dejó de hacerle a la IA preguntas que esta olvida a la mañana siguiente y, en su lugar, logró que mantuviera una wiki viva.
3 comandos ejecutan todo el sistema.
Ingestar: introduce un artículo, un podcast o un PDF de 40 páginas, y Claude lo divide en páginas atómicas enlazadas a todo lo que él ya sabe.
Consultar: pregunta cualquier cosa y responde basándose en sus propias notas, con sus propias palabras, citando sus propias páginas en lugar de adivinar a partir de datos de entrenamiento.
Cerrar el ciclo: cada respuesta se convierte en una nueva nota, y el baúl se afina cada vez que lo toca, multiplicando su valor en lugar de estancarse.
La semana 1 tenía 2000 notas, en su mayoría ruido.
El mes 2 alcanzó las 7.400 con conexiones activándose a toda máquina.
El mes 6 se convirtió en 15.000 notas capaces de debatirle.
Superó a cualquier curso de IA de 2.000 $ que hubiera comprado jamás, porque su sistema ya sabía lo que él sabía.
La estructura de carpetas y el prompt exacto de Claude Code están en la respuesta.
Sígueme si construyes con IA.
Sus notas dejaron de ser un simple almacenamiento.
Empezaron a pensar por sí mismas.
