https://www.youngju.dev/blog 천천히 올바르게. AI Researcher & DevOps Engineer Youngju's tech blog. GPU/CUDA, LLM, MLOps, Kubernetes AI workloads, distributed training, and data engineering. ko fjvbn2003@gmail.com (Youngju Kim) fjvbn2003@gmail.com (Youngju Kim) Sat, 16 May 2026 00:00:00 GMT https://www.youngju.dev/blog/culture/2026-05-16-lean-4-proof-assistants-2026-mathlib-rocq-agda-isabelle-fstar-alphaproof-aimo-deep-dive.en https://www.youngju.dev/blog/culture/2026-05-16-lean-4-proof-assistants-2026-mathlib-rocq-agda-isabelle-fstar-alphaproof-aimo-deep-dive.en In 2026, proof assistants are no longer a secret weapon of academia. Lean 4 has exploded since Leonardo de Moura moved to AWS; mathlib has crossed one million theorems and become a daily tool for first-rank mathematicians including Terence Tao. Coq finished its rebrand to Rocq in March 2024, and Agda, Idris 2, and Isabelle/HOL each hold their own corners. Microsoft's F* has verified the HACL* crypto library that ships in Firefox and the Linux kernel; Liquid Haskell verifies ordinary Haskell with refinement types. On top of all this, AI has arrived — DeepMind's AlphaProof took silver at IMO 2024; Anthropic joined the AIMO prize ($5M) and mathlib search; Terence Tao is formalising his own papers in Lean with Copilot/Claude assistance. Z3 and CVC5 run underneath as the SMT engines; seL4, CompCert, and IronFleet prove that formal verification works in real systems. KAIST and ETRI in Korea, Kyoto University, the University of Tokyo, and AIST in Japan follow. This post puts the whole 2026 proof-assistant ecosystem in one place. Sat, 16 May 2026 00:00:00 GMT fjvbn2003@gmail.com (Youngju Kim) lean-4leanmathlibrocqcoqagdaidris-2isabelle-holfstarliquid-haskellatstheorem-proverproof-assistantformal-methodsalphaproofaimoterence-taoz3cvc5sel4compcert2026deep-diveenglish https://www.youngju.dev/blog/culture/2026-05-16-lean-4-proof-assistants-2026-mathlib-rocq-agda-isabelle-fstar-alphaproof-aimo-deep-dive.ja https://www.youngju.dev/blog/culture/2026-05-16-lean-4-proof-assistants-2026-mathlib-rocq-agda-isabelle-fstar-alphaproof-aimo-deep-dive.ja 2026年、定理証明支援系(proof assistant)はもはや学界の秘密兵器ではない。Lean 4 は Leonardo de Moura が AWS に移って以来爆発的に成長し、mathlib は 100 万定理を超え、Terence Tao を含む一流数学者の日常の道具となった。Coq は 2024 年 3 月に Rocq へリブランドを完了し、Agda・Idris 2・Isabelle/HOL はそれぞれの位置を保つ。Microsoft の F* は HACL* 暗号ライブラリを検証し、Firefox や Linux カーネルに実際に入っている。Liquid Haskell は refinement type で普通のコードを検証する。その上に AI が乗った — DeepMind の AlphaProof は IMO 2024 で銀メダル相当、Anthropic は AIMO($5M)と mathlib 検索に参戦、Terence Tao は自分の論文を Copilot/Claude 補助で Lean に形式化している。Z3・CVC5 SMT ソルバがその下を支え、seL4・CompCert・IronFleet は形式検証が実システムで動くことを証明した。韓国の KAIST・ETRI、日本の京都大学・東京大学・産業技術総合研究所(AIST)もその流れを追う。本記事は 2026 年の定理証明支援系エコシステム全体を一か所に整理する。 Sat, 16 May 2026 00:00:00 GMT fjvbn2003@gmail.com (Youngju Kim) lean-4leanmathlibrocqcoqagdaidris-2isabelle-holfstarliquid-haskellatstheorem-proverproof-assistantformal-methodsalphaproofaimoterence-taoz3cvc5sel4compcert2026deep-dive日本語 https://www.youngju.dev/blog/culture/2026-05-16-lean-4-proof-assistants-2026-mathlib-rocq-agda-isabelle-fstar-alphaproof-aimo-deep-dive https://www.youngju.dev/blog/culture/2026-05-16-lean-4-proof-assistants-2026-mathlib-rocq-agda-isabelle-fstar-alphaproof-aimo-deep-dive 2026년, 정리 증명기(proof assistant)는 더 이상 학계의 비밀 무기가 아니다. Lean 4는 Leonardo de Moura가 AWS로 옮긴 뒤 폭발적으로 성장했고, mathlib는 100만 개 정리를 넘기며 Terence Tao를 비롯한 일급 수학자들의 일상 도구가 되었다. Coq는 2024년 3월 Rocq로 리브랜드를 마쳤고, Agda·Idris 2·Isabelle/HOL은 각자의 자리를 지킨다. Microsoft의 F*는 HACL* 암호 라이브러리를 검증해 Firefox와 Linux 커널에 실제로 들어갔고, Liquid Haskell은 refinement type으로 일반 코드를 검증한다. 그 위로 AI가 올라탔다 — DeepMind AlphaProof는 IMO 2024에서 은메달을, Anthropic은 AIMO ($5M 상금)와 mathlib search에 합류했다. Terence Tao는 Lean과 GitHub Copilot으로 자기 논문을 형식화하고 있다. Z3·CVC5 SMT 솔버가 그 밑단을 받치고, seL4·CompCert·IronFleet는 형식 검증이 실제 시스템에서 작동함을 증명한다. 한국의 KAIST·ETRI, 일본의 京都大学·東京大学·AIST가 그 흐름을 따라간다. 이 글은 2026년 정리 증명기 생태계 전체를 한 자리에 정리한다. Sat, 16 May 2026 00:00:00 GMT fjvbn2003@gmail.com (Youngju Kim) lean-4leanmathlibrocqcoqagdaidris-2isabelle-holfstarliquid-haskellatstheorem-proverproof-assistantformal-methodsalphaproofaimoterence-taoz3cvc5sel4compcert2026deep-dive https://www.youngju.dev/blog/culture/2026-05-16-superhuman-game-ai-2026-stockfish-17-leela-chess-zero-katago-alphazero-muzero-cicero-pluribus-alphastar-shogi-dlshogi-deep-dive.en https://www.youngju.dev/blog/culture/2026-05-16-superhuman-game-ai-2026-stockfish-17-leela-chess-zero-katago-alphazero-muzero-cicero-pluribus-alphastar-shogi-dlshogi-deep-dive.en A complete map of superhuman game AI in 2026. Stockfish 17 (strongest chess), Leela Chess Zero (neural-net chess), Komodo Dragon 3, AlphaZero to MuZero (DeepMind line), Maia (human-like chess), KataGo and Leela Zero (Go), AlphaGo (the Lee Sedol series), Pluribus (6-player poker), Cicero (Diplomacy + language), AlphaStar (StarCraft 2), OpenAI Five (Dota 2), Suphx (Mahjong), AlphaProof + AlphaGeometry (IMO silver), plus lichess/chess.com/ChessBase/Arena/Banksia/NIBBLER UIs, the UCI and XBoard protocols, NCsoft Hancho in Korea, and dlshogi / Yaneura-ou in Japan — who does what, how, and why, all in one place. Sat, 16 May 2026 00:00:00 GMT fjvbn2003@gmail.com (Youngju Kim) game-aichess-enginestockfishleela-chess-zerolc0komodo-dragonalphazeromuzeromaia-chesskatagoleela-zeropluribusciceroalphastaropenai-fivedota-2suphxalphaproofalphageometryalphagolichesschess-comchessbasearenabanksianibblerucixboardlichess-fishnetgogodbonanzayaneura-oudlshogishogi-ai2026deep-diveenglish https://www.youngju.dev/blog/culture/2026-05-16-superhuman-game-ai-2026-stockfish-17-leela-chess-zero-katago-alphazero-muzero-cicero-pluribus-alphastar-shogi-dlshogi-deep-dive.ja https://www.youngju.dev/blog/culture/2026-05-16-superhuman-game-ai-2026-stockfish-17-leela-chess-zero-katago-alphazero-muzero-cicero-pluribus-alphastar-shogi-dlshogi-deep-dive.ja 2026年の超人間ゲームAIを一枚の地図にまとめる。Stockfish 17（チェス最強）、Leela Chess Zero（NNベース）、Komodo Dragon 3、AlphaZeroからMuZeroへ続くDeepMind系、Maia（人間らしいチェス）、KataGoとLeela Zero（囲碁）、AlphaGo（李世乭との対局）、Pluribus（6人ポーカー）、Cicero（Diplomacy + 言語）、AlphaStar（StarCraft 2）、OpenAI Five（Dota 2）、Suphx（麻雀）、AlphaProof + AlphaGeometry（IMO銀メダル）。lichess・chess.com・ChessBase・Arena・Banksia・NIBBLERといったUI、UCIとXBoardプロトコル、韓国のハンドル(韓乭)と日本のdlshogi・やねうら王まで — 誰が・何を・どうやっているかを一気通貫で読める。 Sat, 16 May 2026 00:00:00 GMT fjvbn2003@gmail.com (Youngju Kim) game-aichess-enginestockfishleela-chess-zerolc0komodo-dragonalphazeromuzeromaia-chesskatagoleela-zeropluribusciceroalphastaropenai-fivedota-2suphxalphaproofalphageometryalphagolichesschess-comchessbasearenabanksianibblerucixboardlichess-fishnetgogodbonanzayaneura-oudlshogishogi-ai2026deep-dive日本語 https://www.youngju.dev/blog/culture/2026-05-16-superhuman-game-ai-2026-stockfish-17-leela-chess-zero-katago-alphazero-muzero-cicero-pluribus-alphastar-shogi-dlshogi-deep-dive https://www.youngju.dev/blog/culture/2026-05-16-superhuman-game-ai-2026-stockfish-17-leela-chess-zero-katago-alphazero-muzero-cicero-pluribus-alphastar-shogi-dlshogi-deep-dive 2026년 초인간 게임 AI를 한 글로 정리한다. Stockfish 17(체스 최강), Leela Chess Zero(신경망), Komodo Dragon 3, AlphaZero에서 MuZero로 이어진 DeepMind 계열, Maia(인간 같은 체스), KataGo와 Leela Zero(바둑), AlphaGo(이세돌), Pluribus(6명 포커), Cicero(Diplomacy + 언어), AlphaStar(StarCraft 2), OpenAI Five(Dota 2), Suphx(마작), AlphaProof + AlphaGeometry(IMO 은메달), 그리고 lichess/chess.com/ChessBase/Arena/Banksia/NIBBLER 같은 UI, UCI와 XBoard 프로토콜, 한국의 한돌과 일본의 dlshogi·やねうら王까지 — 누가 무엇을 어떻게 하는지를 한 번에. Sat, 16 May 2026 00:00:00 GMT fjvbn2003@gmail.com (Youngju Kim) game-aichess-enginestockfishleela-chess-zerolc0komodo-dragonalphazeromuzeromaia-chesskatagoleela-zeropluribusciceroalphastaropenai-fivedota-2suphxalphaproofalphageometryalphagolichesschess-comchessbasearenabanksianibblerucixboardlichess-fishnetgogodbonanzayaneura-oudlshogishogi-ai2026deep-dive