https://www.youngju.dev/blog 천천히 올바르게. AI Researcher & DevOps Engineer Youngju's blog. GPU/CUDA, LLM, MLOps, Kubernetes AI workloads, and data engineering — plus mindset essays on confidence, routines, health, and sport psychology. ko fjvbn2003@gmail.com (Youngju Kim) fjvbn2003@gmail.com (Youngju Kim) Sun, 09 Aug 2026 00:00:00 GMT https://www.youngju.dev/blog/ai/2026-08-09-llms-cant-jump-abduction-gap.en https://www.youngju.dev/blog/ai/2026-08-09-llms-cant-jump-abduction-gap.en The ICML 2026 position paper Position: LLMs can not jump argues that generative AI has mastered induction and is rapidly conquering deduction, yet remains structurally unable to reach abduction, the act of producing a new explanatory hypothesis. Rather than summarize that argument, this post asks how the systems we build right now should change if we assume it is true: where the hypothesis space should be supplied from, why proposal and falsification have to be separated, and how the fact that this is a position paper should shape the way you read it. Sun, 09 Aug 2026 00:00:00 GMT fjvbn2003@gmail.com (Youngju Kim) aillmreasoningabductionresearch https://www.youngju.dev/blog/ai/2026-08-09-llms-cant-jump-abduction-gap.ja https://www.youngju.dev/blog/ai/2026-08-09-llms-cant-jump-abduction-gap.ja ICML 2026の立場論文 Position: LLMs can not jump は、生成AIが帰納をすでにマスターし演繹も急速に征服しつつある一方で、新しい説明仮説を作り出すアブダクションには構造的に到達できないと主張します。この記事はその主張を要約する代わりに、それが事実だと仮定したときに、いま私たちが作っているシステムの設計をどう変えるべきかを扱います。仮説空間をどこから供給するのか、提案と反証をなぜ分離すべきなのか、そしてこの論文が立場論文であるという事実をどう踏まえて読むべきなのか、です。 Sun, 09 Aug 2026 00:00:00 GMT fjvbn2003@gmail.com (Youngju Kim) aillmreasoningabductionresearch https://www.youngju.dev/blog/ai/2026-08-09-llms-cant-jump-abduction-gap https://www.youngju.dev/blog/ai/2026-08-09-llms-cant-jump-abduction-gap ICML 2026 입장 논문 Position: LLMs can not jump은 생성형 AI가 귀납을 마스터했고 연역을 빠르게 정복하는 중이지만, 새로운 설명 가설을 만들어 내는 귀추에는 구조적으로 도달하지 못한다고 주장합니다. 이 글은 그 주장을 요약하는 대신, 그것이 사실이라고 가정했을 때 우리가 지금 만드는 시스템의 설계를 어떻게 바꿔야 하는지를 다룹니다. 가설 공간을 어디서 공급할 것인가, 제안과 반증을 왜 분리해야 하는가, 그리고 논문이 입장 논문이라는 사실을 어떻게 감안해서 읽어야 하는가입니다. Sun, 09 Aug 2026 00:00:00 GMT fjvbn2003@gmail.com (Youngju Kim) aillmreasoningabductionresearch https://www.youngju.dev/blog/ai/2026-08-09-llms-cant-jump-abduction-gap.zh https://www.youngju.dev/blog/ai/2026-08-09-llms-cant-jump-abduction-gap.zh ICML 2026 的立场论文 Position: LLMs can not jump 主张，生成式 AI 已经掌握了归纳，也正在快速攻下演绎，但对于造出新解释性假说的溯因推理，它在结构上根本到不了。本文不去复述那套论证，而是去问：假定它成立的话，我们此刻正在做的系统，设计上该改哪里。假说空间从哪里供给，为什么必须把提出与反驳拆开，以及「这是一篇立场论文」这个事实该如何影响你的读法。 Sun, 09 Aug 2026 00:00:00 GMT fjvbn2003@gmail.com (Youngju Kim) aillmreasoningabductionresearch