EP 87. '딸깍'의 시대, 슬픔과 기쁨 사이

On a Saturday morning, February 21, 2026, Chester Roh and Seungjoon Choi unpack the overwhelming pace of recent AI model and agent trends, the FOMO they bring, and the emotional toll sometimes called "AI depression." They cover the latest developments including METR's Claude Opus 4.6 benchmarks, GPT-5.2, and Gemini 3.1 Pro, while exploring the question: "What's the minimum you need to know to head in the right direction, even when you don't know everything?" (MVK — Minimum Viable Knowledge). From harness/scaffolding engineering and the human-in-the-loop perspective, they arrive at the message that "code isn't the goal — meaningful software is," and reflect on the enduring value of vision, judgment, and taste for engineers. They also share the joy of rediscovering hands-on fun by coding along with Andrej Karpathy's microgpt and experimenting with automatic differentiation for creative coding — searching together for healthy ways to learn and explore amid rapid change. 00:00:00 Intro: Anxiety in the Age of One Click 00:01:15 MVK — Heading the Right Way Even When You Don't Know 00:01:47 METR Benchmark Saturation: Claude Opus 4.6's 14-Hour Run 00:03:40 Gemini 3.1 Pro and the Never-Ending Stream of New Models 00:05:11 Chris Lattner's Deep Review of Claude's C Compiler 00:08:35 Andrej Karpathy's March of Nines: Between 90% and 99.99% 00:10:12 Everything as a Search Problem: Science and Drug Discovery 00:11:07 OpenAI's Theoretical Physics Case and the Centaur Era 00:14:14 Are Model Capabilities Interpolation or Extrapolation? 00:18:19 The Reality of Agentic Coding: The Limits of the Ralph Loop 00:21:41 Harness Engineering 00:25:11 The FOMO Industry and AI Depression 00:28:05 In the One-Click World, Good Software Will Only Grow 00:29:48 Rediscovering Fun with Andrej Karpathy's microgpt 00:31:34 Creative Applications of Automatic Differentiation and the Joy of Learning 00:37:11 Process Over Results: Rediscovering the Fun of Learning and Taste 00:42:33 Using AI Healthily: The Mechanics of AI-Driven Burnout 00:43:54 Gilbert Strang's Linear Algebra and the Discipline of Deep Study 00:46:23 The Difference Between 99 and 99.99 00:50:08 20K Subscribers — Thank You 📝 Read the full transcript: https://aifrontier.kr/en/episodes/ep87