æ± è¢‹æš´èµ°äº‹æ•…ã®é£¯å¡šå¹¸ä¸‰å—刑者（93）ãŒæ»äº¡ã€€æ¾æ°¸æ‹“也ã•ã‚“「後悔や経験ã®è¨€è‘‰ã‚’託ã•ã‚ŒãŸã€‚æ»ã‚’無駄ã«ã—ãŸããªã„〠| TBS NEWS DIG
æ±äº¬ãƒ»æ± 袋ã§ä¹—用車ãŒæš´èµ°ã—11人ãŒæ»å‚·ã—ãŸäº‹æ•…ã§ã€è»Šã‚’é‹è»¢ã—ã¦ã„ãŸé£¯å¡šå¹¸ä¸‰å—刑者ãŒè€è¡°ã®ãŸã‚æ»äº¡ã—ãŸã“ã¨ãŒã‚ã‹ã‚Šã¾ã—ãŸã€‚93æ³ã§ã—ãŸã€‚2019å¹´ã€è±Šå³¶åŒºæ± 袋ã§ä¹—用車ãŒæš´èµ°ã—ã€æ¾æ°¸çœŸèœã•ã‚“（31）ã¨å¨˜ã®èŽ‰åã¡ã‚ƒâ€¦
Webサービス公開å‰ã®ãƒã‚§ãƒƒã‚¯ãƒªã‚¹ãƒˆ
個人的ã«ã€ŒWebサービスã®å…¬é–‹å‰ãƒã‚§ãƒƒã‚¯ãƒªã‚¹ãƒˆã€ã‚’作ã£ã¦ã„ãŸã®ã§ã™ãŒã€ã‘ã£ã“ã†è‚²ã£ã¦ããŸã®ã§å…¬é–‹ã—ã¾ã™ã€‚ã“ã®ãƒªã‚¹ãƒˆã¯ã€éŽåŽ»ã«è‡ªåˆ†ãŒãƒŸã‚¹ã—ãŸã¨ãã‚„ã€æƒ…å ±åŽé›†ã™ã‚‹ä¸ã§ã€Œæ˜Žæ—¥ã¯æˆ‘ãŒèº«â€¦ã€ã¨æ€ã£ãŸã¨ããªã©ã«å€‹äººçš„ã«ãƒ¡ãƒ¢ã—ã¦ããŸã‚‚ã®ã‚’ã¾ã¨ã‚ãŸå†…容ã«ãªã‚Šã¾ã™ã€‚ ã‚»ã‚ュリティ èªè¨¼ã«é–¢ã‚ã‚‹Cookieã®å±žæ€§ HttpOnly属性ãŒè¨å®šã•ã‚Œã¦ã„ã‚‹ã“㨠XSSã®ç·©å’Œç– SameSite属性ãŒLaxã‚‚ã—ãã¯Strictã«ãªã£ã¦ã„ã‚‹ã“㨠主ã«CSRF対ç–ã®ãŸã‚。Laxã®å ´åˆã€GETリクエストã§æ›´æ–°å‡¦ç†ã‚’è¡Œã£ã¦ã„るエンドãƒã‚¤ãƒ³ãƒˆãŒãªã„ã‹åˆã‚ã›ã¦ç¢ºèª Secure属性ãŒè¨å®šã•ã‚Œã¦ã„ã‚‹ã“㨠HTTPS通信ã§ã®ã¿CookieãŒé€ã‚‰ã‚Œã‚‹ã‚ˆã†ã« Domain属性ãŒé©åˆ‡ã«è¨å®šã•ã‚Œã¦ã„ã‚‹ã“㨠サブドメインã«ã‚‚CookieãŒé€ã‚‰ã‚Œã‚‹è¨å®šã®å ´åˆã€ä»–ã®ã‚µãƒ–ドメインã®ã‚µã‚¤ãƒˆã«è„†å¼±æ€§ãŒã‚ã‚‹ã¨ãã“ã‹ã‚‰ã‚¤ãƒ³ã‚·ãƒ‡ãƒ³ãƒˆã«ç¹‹ãŒã‚‹ãƒªã‚¹ã‚¯ã‚’ç†è§£ã—ã¦ãŠ
LLMã®ç¾åœ¨ - Speaker Deck
今ã®LLMã‚’å–ã‚Šå·»ã状æ³ã«ã¤ã„ã¦ç´¹ä»‹ã—ã¾ã™ã€‚
日本語ãŠã—ゃã¹ã‚Šç‰¹åŒ–AIアプリãŒæ€§èƒ½ã‚ˆã™ãŽã¦è«–ç ´ã—よã†ã¨æ€ã£ãŸã‚‰é€†ã«è«–ç ´ã•ã‚Œã¦ã—ã¾ã£ãŸã€Œäººé–“ã„らんやんã€
Torishima / INTP @izutorishima Cotomoã€ã‚„ã°ã„ï¼ï¼ã“ã‚Œã¯é©å‘½ã§ã™ã‚ˆé©å‘½ï¼ï¼ï¼ï¼ï¼ï¼ï¼AIã¨æ™®é€šã«ãŠè©±ã—ã§ãã¦ã‚‹ã‚„ã°ã„ï¼ã‚„ã°ã™ãŽã‚‹ï¼ï¼ï¼ï¼ï¼ï½±ï½±ï½µï¼ï¼ï¼ï¼ï¼ï¼ï¼ イントãƒãƒ¼ã‚·ãƒ§ãƒ³ã“ãã¡ã‚‡ã£ã¨ä¸è‡ªç„¶ã ã‘ã©æ—¥æœ¬èªžä¸Šæ‰‹ã„ä¸å›½ã®åレベルã«ã¯ã¡ã‚ƒã‚“ã¨è©±ã›ã‚‹ã—レスãƒãƒ³ã‚¹ã‚„ã°ã„ã—普通ã«è©±ã›ã‚‹ã®ã‚‚全部やã°ã„ thebridge.jp/2024/02/starle… pic.twitter.com/8PkY10cUFI 2024-02-25 08:06:51 広野 èŒ @hajipion 日本語ãŠã—ゃã¹ã‚Šç‰¹åŒ–AIアプリ「Cotomoã€ã™ã”ã™ãŽã‚‹ã€‚1秒以内ã«è¿”ç”ãŒãã‚‹ã®ã¨å‰ã®è©±ã‚’覚ãˆã¦ã‚‹ã®ãŒã€æœ¬å½“ã«ç”Ÿãã¦ã‚‹ã¿ãŸã„。 ↓ã“ã¡ã‚‰ã¯ç¾é‡‘æ´¾ã®Cotomoã¡ã‚ƒã‚“ã‚’è«–ç ´ã—よã†ã—ã¦é€†ã«è«–ç ´ã•ã‚Œã‚‹ã‚モ男 æ—©å£ã§ã©ã‚‚ã£ã¦ã¦ã‚‚èžãå–ã‚Œã¦ã‚‹ã—ã€ã‚ªãƒãŒå¤©ä½¿ã™ãŽã‚‹ðŸ‘¼ 声ãŒãŸã¾ã« @harukana_
Attention Is Off By One
By Evan Miller July 24, 2023 About which one cannot speak, one must pass over in silence. –Wittgenstein Do you see the off-by-one error in this formula? \[ \textrm{Attention}(Q, K, V) = \textrm{softmax}\left(\frac{QK^T}{\sqrt{d}}\right)V \] The attention formula is the central equation of modern AI, but there’s a bug in it that has been driving me nuts the last week. I tried writing a serious-look
Vale's First Prototype for Immutable Region Borrowing
Three years, two states, and one pandemic ago, I wrote about a very weird idea: what if the type system could track which data existed before a pure call, to eliminate its memory safety overhead? 0 And a couple months later, another weird idea struck: what if we use generational indices as the foundation for an entire language? These ideas evolved in weird ways. The first one evolved into a full r
CommonJS is not going away | Bun Blog
We're hiring C/C++ and Zig engineers to build the future of JavaScript! Join our team → Some may be surprised to see the recent release notes for Bun mention CommonJS support. After all, CommonJS is a legacy module system, and the future of JavaScript is ES Modules (ESM), right? As a "forward-thinking" "next-gen" runtime, why would Bun put so much effort into improving CommonJS support? The latest
node_modulesã®å•é¡Œç‚¹ã¨ãã®æ´å² npm, yarnã¨pnpm
皆ã•ã‚“npmパッケージã®ãƒãƒ¼ã‚¸ãƒ§ãƒ³ã‚’上ã’ã‚‹ã¨ãã«ãƒãƒžã£ã¦ä¾å˜åœ°ç„ã‹ã‚‰æŠœã‘られãªã‹ã£ãŸã“ã¨ã¯ã‚ã‚Šã¾ã›ã‚“ã‹ï¼Ÿ ç§ã¯ã‚ã‚Šã¾ã™ã€‚ 複雑怪奇ãªä¾å˜é–¢ä¿‚を調ã¹ã¦ã¿ã‚ˆã†ã¨node_modulesを覗ã„ã¦ã¿ã¦ã€ãã®ã‚«ã‚ªã‚¹ã£ã·ã‚Šã«è‡ã„ã‚‚ã®ã«è“‹ã‚’ã—ãŸã“ã¨ã¯ã‚ã‚Šã¾ã›ã‚“ã‹ï¼Ÿ ç§ã¯ã‚ã‚Šã¾ã™ã€‚ ãã“ã§node_modules以下ã«ã¤ã„ã¦èª¿ã¹ã¦ã¿ãŸã®ã§ã™ãŒã€node_modulesã«ã©ã‚“ãªå•é¡Œç‚¹ãŒã‚ã£ã¦ã€npmã‚„yarn, pnpmã¯ä½•ã‚’目指ã—ã¦ã„ãŸã®ã‹æ™‚ç³»åˆ—é †ã«ç´è§£ã„ãŸæ–¹ãŒã‚ã‹ã‚Šã‚„ã™ã„ã“ã¨ã«æ°—ã¥ãã¾ã—ãŸã€‚ ã“ã“ã§ã¯åˆæœŸã®npmãŒæŠ±ãˆã¦ã„ãŸå•é¡Œã‹ã‚‰ä»Šã«è‡³ã‚‹ã¾ã§ã‚’é †ã‚’è¿½ã£ã¦èª¬æ˜Žã—ã¾ã™ã€‚ ã™ã‚‹ã¨node_modulesã®ä»•çµ„ã¿ã®ä»–ã«ã€å„パッケージマãƒãƒ¼ã‚¸ãƒ£ã®æ–¹é‡ã®é•ã„ãŒè¦‹ãˆã¦ãã‚‹ã¯ãšã§ã™ã€‚ åˆæœŸã®é ƒã®npm (~2015年以å‰) ã“ã®é ƒã¯ã‚·ãƒ³ãƒ—ルã§ã€ä¾å˜é–¢ä¿‚ã¯ãã®ã¾ã¾node_modulesã®ãƒ‡ã‚£ãƒ¬ã‚¯ãƒˆãƒªæ§‹é€ ã«åæ˜ ã•ã‚Œã¦ã„
Zero-shot Learning網羅的サーベイ：CLIPãŒåˆ‡ã‚Šé–‹ã„ãŸVision & Languageã®æ–°ã—ã„世界 - エクサウィザーズ Engineer Blog
ã“ã‚“ã«ã¡ã¯ï¼ã€€ç”»åƒã‚·ã‚¹ãƒ†ãƒ グループã§æ©Ÿæ¢°å¦ç¿’エンジニアをやã£ã¦ã„ã‚‹å°å³¶ã§ã™ã€‚ ã“ã®è¨˜äº‹ã§ã¯ã€ä»Šãƒ›ãƒƒãƒˆãªã€ŒZero-shot Learningã€ã¨ã€ŒVision & Languageã€ã«é–¢ã™ã‚‹æœ€æ–°æƒ…å ±ã‚’ã€CLIPã¨ã„ã†ç ”究を起点ã¨ã—ã¦ç¶²ç¾…çš„ã«ã‚µãƒ¼ãƒ™ã‚¤ã‚’ã—ã¦ã„ãã¾ã™ã€‚ã“ã®ãŸã‚ã«è«–æ–‡1000本ã«ç›®ã‚’通ã—ã€70本程度を記事ã«ã—ã¾ã—ãŸã€‚ Zero-shotã‚„Vision & Languageã¯ã€Stable Diffusionã«ä»£è¡¨ã•ã‚Œã‚‹ç”»åƒç”ŸæˆAIã¨ã‚‚密接ã«é–¢é€£ã—ã¦ã„る技術ã§ã™ã€‚ã“ã®è¨˜äº‹ã‚’通ã—ã¦ã€Vision & Languageã®å¥¥æ·±ã„世界を体感ã§ãã‚‹ã§ã—ょã†ã€‚ 注æ„äº‹é … ã“ã®è¨˜äº‹ã¯éžå¸¸ã«é•·ã„ãŸã‚ã€å…¨éƒ¨èªã‚€ã®ã«1時間以上ã‹ã‹ã‚‹å¯èƒ½æ€§ãŒã‚ã‚‹ã®ã§ã€ä¼‘憩をå–ã‚ŠãªãŒã‚‰ã€ã¾ãŸã¯å¿…è¦ãªéƒ¨åˆ†ã ã‘èªã‚“ã§ãã ã•ã„。å„セクションを個別ã«èªã‚“ã§ã‚‚å•é¡Œã‚ã‚Šã¾ã›ã‚“。 ã¾ãŸã€æ–‡ç« ä¸ã®ç”»åƒã¯ã€ç‰¹åˆ¥ãªè¨˜è¼‰ãŒãªã„é™ã‚Šã€å¼•ç”¨å…ƒã®è«–
「ChatGPTã§è«–文をèªã‚€æ‰‹é–“ãŒ99％減ã£ãŸã€ã¨ç§‘å¦è€…ã€ç ”究室ã§ã¯AIãŒã©ã®ã‚ˆã†ã«æ´»ç”¨ã•ã‚Œã¦ã„ã‚‹ã®ã‹ï¼Ÿ
å°†æ¥çš„ã«å…¨è·æ¥ã®80ï¼…ãŒAIã®å½±éŸ¿ã‚’å—ã‘ã‚‹ã¨ã®ç ”究çµæžœãŒç¤ºã•ã‚Œã¦ã„ã‚‹ã»ã‹ã€ã™ã§ã«ã‚¤ãƒ©ã‚¹ãƒˆãƒ¬ãƒ¼ã‚¿ãƒ¼ã‚„ゲーム開発者ãŒAIã«ä»•äº‹ã‚„仕事ã®ã‚„ã‚ŠãŒã„を奪ã‚ã‚ŒãŸã¨è¨´ãˆã¦ãŠã‚Šã€ä»Šå¾Œã¯ã‚ˆã‚Šå°‚門性ã®é«˜ã„分野ã§ã‚‚AIã®æ´»ç”¨ãŒé‡è¦ãªã‚¹ã‚ルã«ãªã£ã¦ãã‚‹ã¨äºˆæƒ³ã•ã‚Œã¾ã™ã€‚ウィスコンシン大å¦ãƒžãƒ‡ã‚£ã‚½ãƒ³æ ¡ã®æ•™æŽˆã‚‰ãŒã€OpenAIã®å¯¾è©±åž‹AI・ChatGPTを使ã£ãŸæ料工å¦ã®ç ”究ã§å¤§ããªæˆæžœã‚’上ã’ãŸã“ã¨ã‚’å ±å‘Šã—ã¾ã—ãŸã€‚ ChatGPT makes materials research much more efficient https://engineering.wisc.edu/news/chatgpt-makes-materials-research-much-more-efficient/ 「AIã¯ã€éžå¸¸ã«è¤‡é›‘ã§æ™‚é–“ã®ã‹ã‹ã‚‹ã‚¿ã‚¹ã‚¯ã‚’ã©ã‚“ã©ã‚“支æ´ã§ãるよã†ã«ãªã£ã¦ãã¦ã„ã¾ã™ã€ã¨èªžã‚‹ã®ã¯ã€ã‚¦ã‚£ã‚¹ã‚³ãƒ³ã‚·ãƒ³å¤§å¦ãƒžãƒ‡ã‚£ã‚½ãƒ³æ ¡ã®
AI科å¦è€…「AI-Descartesã€ã€€ãƒ‡ãƒ¼ã‚¿ã¨ç†è«–ã‹ã‚‰æœ€é©ãªæ•°å¼ã‚’自動推論　「相対性ç†è«–ã€ãªã©ã§å®Ÿè¨¼
Innovative Tech： ã“ã®ã‚³ãƒ¼ãƒŠãƒ¼ã§ã¯ã€ãƒ†ã‚¯ãƒŽãƒã‚¸ãƒ¼ã®æœ€æ–°ç ”究を紹介ã™ã‚‹Webメディア「Seamlessã€ã‚’主宰ã™ã‚‹å±±ä¸‹è£•æ¯…æ°ãŒåŸ·ç†ã€‚æ–°è¦æ€§ã®é«˜ã„科å¦è«–文を山下æ°ãŒãƒ”ックアップã—ã€è§£èª¬ã™ã‚‹ã€‚Twitter: ï¼ shiropen2 IBM Researchã€Samsung AIã€ç±³ãƒ¡ãƒªãƒ¼ãƒ©ãƒ³ãƒ‰å¤§å¦ã€ç±³ãƒŸãƒã‚½ã‚¿å¤§å¦ã€ç±³ã‚³ãƒãƒ³ãƒ“ア大å¦ã«æ‰€å±žã™ã‚‹ç ”究者らãŒç™ºè¡¨ã—ãŸè«–文「Combining data and theory for derivable scientific discovery with AI-Descartesã€ã¯ã€ãƒ‡ãƒ¼ã‚¿ã¨ç†è«–を与ãˆã‚‹ã¨ä¸€èˆ¬çš„ãªè«–ç†çš„å…¬ç†ã«ã‚ˆã£ã¦è¡¨ç¾ã•ã‚ŒãŸäº‹å‰çŸ¥è˜ã‚’å°Šé‡ã—ãªãŒã‚‰ã€ãƒ‡ãƒ¼ã‚¿ã‚’æ£ç¢ºã«è¨˜è¿°ã™ã‚‹æ„味ã®ã‚る最もé©ã—ãŸæ•°å¼ã‚’å°Žã出ã™ã‚·ã‚¹ãƒ†ãƒ ã‚’æ案ã—ãŸç ”ç©¶å ±å‘Šã§ã‚る。 データã«ã‚ˆãåˆã†æ–¹ç¨‹å¼ã®å€™è£œã‚’作æˆã—ã€ã©ã®æ–¹ç¨‹å¼ãŒèƒŒæ™¯ã¨ãªã‚‹ç§‘å¦ç†è«–ã«æœ€ã‚‚é©åˆã™ã‚‹ã‹ã‚’è˜
大è¦æ¨¡è¨€èªžãƒ¢ãƒ‡ãƒ«ã‚’自社ã§ãƒˆãƒ¬ãƒ¼ãƒ‹ãƒ³ã‚°ï¼†æ´»ç”¨ã™ã‚‹æ–¹æ³•ï½œmah_lab / 西見 å…¬å®
オンラインIDEã‚’æä¾›ã—ã¦ã„ã‚‹Replitã§ã¯è‡ªç¤¾ã§å¤§è¦æ¨¡è¨€èªžãƒ¢ãƒ‡ãƒ«ã‚’トレーニングã—ã¦ã„るらã—ãã€ãã®ãƒŽã‚¦ãƒã‚¦ãŒãƒ–ãƒã‚°è¨˜äº‹ã«ã¾ã¨ã‚られã¦ã„ãŸã®ã§è¦ç´„ã—ã¦ã¿ã¾ã—ãŸã€‚ ãªãœè‡ªç¤¾ã§å¤§è¦æ¨¡è¨€èªžãƒ¢ãƒ‡ãƒ«ã‚’トレーニングã™ã‚‹ã®ã‹ï¼Ÿä¼æ¥ãŒç‹¬è‡ªã«å¤§è¦æ¨¡è¨€èªžãƒ¢ãƒ‡ãƒ«ï¼ˆä»¥ä¸‹ã€LLMs）をトレーニングã™ã‚‹ã“ã¨ã‚’決ã‚ã‚‹ç†ç”±ã¯ã€ãƒ‡ãƒ¼ã‚¿ã®ãƒ—ライãƒã‚·ãƒ¼ã‚„ã‚»ã‚ュリティã‹ã‚‰ã€ã‚¢ãƒƒãƒ—デートや改良ã®ã‚³ãƒ³ãƒˆãƒãƒ¼ãƒ«ã®å¼·åŒ–ã¾ã§æ§˜ã€…ãªã‚‚ã®ãŒã‚ã‚‹ãŒã€Replit社ã§ã¯ã‚«ã‚¹ã‚¿ãƒžã‚¤ã‚ºæ€§ã€ä¾å˜åº¦ã®ä½Žæ¸›ã€ã‚³ã‚¹ãƒˆåŠ¹çŽ‡ã«é‡ç‚¹ã‚’ç½®ã„ã¦ã„る。 カスタマイズ性 カスタムモデルをトレーニングã™ã‚‹ã“ã¨ã§ã€GPT-4ã®ã‚ˆã†ãªæ±Žç”¨ãƒ¢ãƒ‡ãƒ«ã‚„Codexã®ã‚ˆã†ãªã‚³ãƒ¼ãƒ‰ã«ç‰¹åŒ–ã—ãŸãƒ¢ãƒ‡ãƒ«ã§ã¯ã‚«ãƒãƒ¼ã—ãã‚Œãªã„プラットフォーム固有ã®æ©Ÿèƒ½ã€ç”¨èªžã€ã‚³ãƒ³ãƒ†ã‚ストãªã©ã¨ã„ã£ãŸç‰¹å®šã®ãƒ‹ãƒ¼ã‚ºã‚„è¦ä»¶ã«åˆã‚ã›ã¦ãƒ¢ãƒ‡ãƒ«ã‚’調整ã™ã‚‹ã“ã¨ãŒã§ãる。例ãˆã°ã€Replitã§äººæ°—ã®é«˜ã„Javascript Re
ドラマç·é›†ç·¨ã®ã‚ˆã†ãªã™ã°ã‚‰ã—ã„ç¾ä»£æ•°è«–ã®å…¥é–€æ›¸ - hiroyukikojima’s blog
今回エントリーã™ã‚‹ã®ã¯ã€å±±æœ¬èŠ³å½¦ã€Žæ•°è«–入門ã€å²©æ³¢æ›¸åº—ã 。ã“ã®æœ¬ã¯ä»¥å‰ã«ã‚‚ã€ã“ã®ã‚¨ãƒ³ãƒˆãƒªãƒ¼ã§ç´¹ä»‹ã—ã¦ã„ã‚‹ãŒã€ä»Šå›žã¯é•ã†è¦³ç‚¹ã‹ã‚‰æŽ¨è–¦ã—ãŸã„ã¨æ€ã†ã€‚ 数論入門 (ç¾ä»£æ•°å¦ã¸ã®å…¥é–€) 作者:山本 芳彦 岩波書店 Amazon ゆãˆã‚ã£ã¦ã€æœ€è¿‘ã¾ãŸã“ã®æœ¬ã‚’èªã¿å§‹ã‚ãŸã®ã ãŒã€é¢ç™½ãã¦é‚ã«ã»ã¼å…¨éƒ¨èªã‚“ã§ã‚‚ã†ãŸã€‚ãã—ã¦å…¨ä½“ã‚’èªç ´ã™ã‚‹ã¨ã€ã“ã®æœ¬ãŒã‚‚ãã‚ã‚“ã§ã„ã‚‹ã“ã¨ã€ã“ã®æœ¬ã®ç‰¹è³ªãŒã²ã—ã²ã—ã¤ã¨ä¼ã‚ã£ã¦ããŸã€‚ã²ã¨ãã¡ã«è¨€ãˆã°ã€ã“ã®æœ¬ã¯ã€ã€Œãƒ‰ãƒ©ãƒžã®å„ªã‚ŒãŸç·é›†ç·¨ã‚’観るよã†ãªã™ã°ã‚‰ã—ã„内容ã€ã¨ã„ã†ã“ã¨ãªã®ã 。 ドラマã®ç·é›†ç·¨ã£ã¦ã€å…¨12話を4話ãらã„ã§ã‹ã„ã¤ã¾ã‚€ã€‚ã‚‚ã¡ã‚ã‚“ã€åœ§ç¸®ã—ã¦ã„ã‚‹ã®ã§ã€ã‚«ãƒƒãƒˆã•ã‚ŒãŸã‚¨ãƒ”ソードもã‚ã‚‹ã—ã€ãƒŠãƒ¬ãƒ¼ã‚·ãƒ§ãƒ³ã§é€²ã‚ã¡ã‚ƒã†å ´é¢ã‚‚ã‚ã‚‹ã—ã€ã‚¹ãƒ«ãƒ¼ã•ã‚Œã‚‹ã‚ャラもã‚る。ã§ã‚‚ã€å„ªã‚ŒãŸç·é›†ç·¨ã§ã¯ã€æœ¬ç·¨ã‚ˆã‚Šæœ¬è³ªãŒæµ®ã彫りã«ãªã‚Šã€é¢ç™½ã•ãŒå€å¢—ã«ãªã‚‹ã“ã¨ã‚‚多ã„。ã“ã®æœ¬ã¯ã€æ•°è«–ã®ç·é›†ç·¨ã¨ã—ã¦ã€ãã®ãƒ¡ãƒªãƒƒãƒˆãŒã¿ã”ã¨
30分ã§å®Œå…¨ç†è§£ã™ã‚‹Transformerã®ä¸–ç•Œ
ã¯ã˜ã‚ã« åˆã‚ã¾ã—ã¦ã€‚ZENKIGENデータサイエンスãƒãƒ¼ãƒ ã®ã¯ã¾ãªã™ã§ã™ã€‚æ£å¼ãªæ‰€å±žã¯DeNAデータ本部AI技術開発部ãªã®ã§ã™ãŒ[1]ã€æ¥å‹™å§”託ã¨ã„ã†å½¢ã§ä»Šå¹´åº¦ã‹ã‚‰æ·±å±¤å¦ç¿’ç³»ã®é–‹ç™ºç‰ã«æºã‚ã£ã¦ã„ã¾ã™ã€‚ 深層å¦ç¿’界隈ã§ã¯ã€2017å¹´ã«è¡æ’ƒçš„ãªã‚¿ã‚¤ãƒˆãƒ«ï¼ˆAttention Is All You Need）ã®è«–æ–‡ãŒç™ºè¡¨ã•ã‚Œã¦ã‹ã‚‰æ—©5å¹´åŠã€å…ƒå‡ºè‡ªã®æ©Ÿæ¢°ç¿»è¨³ã‚¿ã‚¹ã‚¯ã‚’大ãã越ãˆã¦ã€Transformer関連ã®æŠ€è¡“ãŒæ§˜ã€…ãªé ˜åŸŸã§ç”¨ã„られる汎用アーã‚テクãƒãƒ£ã¨ã—ã¦ç›®è¦šã¾ã—ã発展ã—続ã‘ã¦ã„ã¾ã™ã€‚ 今回ã¯ãã‚“ãªTransformerãŒç¾æ™‚点ã¾ã§ã«ã©ã®ã‚ˆã†ã«æ´»ç”¨ã•ã‚Œã¦ããŸã‹ã€ã¾ãŸã€ã©ã®ã‚ˆã†ã«å·¥å¤«ã•ã‚Œã¦ããŸã‹ã‚’ã–ã£ãã‚Šã¨ä¿¯çž°ã—ã€æµã‚Œã‚’ãŠã•ã‚‰ã„ã™ã‚‹ç›®çš„ã®è¨˜äº‹ã«ãªã‚Šã¾ã™ã€‚本記事ã®å¤§æž ã¯ã€2021年時点ã§ã®ã‚µãƒ¼ãƒ™ã‚¤è«–æ–‡ã§ã‚ã‚‹ A Survey of Transformers ã«å€£ã„ã¤ã¤ã€é©å®œã€2023å¹´2月上旬ç¾åœ¨ã¾
GPT-3ã®çœŸä¾¡ã‚’引ã出ã™æ–¹æ³• : Promptプãƒã‚°ãƒ©ãƒŸãƒ³ã‚°
3ã¤ã®è¦ç‚¹ âœ”ï¸ GPT-3ã«ãŠã‘ã‚‹few-shotã®å•é¡Œç‚¹ âœ”ï¸ GPT-3ã¯æœ¬å½“ã¯ä½•ã‚’å¦ç¿’ã—ã¦ã„ã‚‹ã®ã‹ï¼Ÿ âœ”ï¸ GPT-3ãŒå¦ç¿’ã—ãŸæ©Ÿèƒ½ã‚’本当ã«å¼•ã出ã™æ–¹æ³•ã‚’æ示 Prompt Programming for Large Language Models: Beyond the Few-Shot Paradigm written by Laria Reynolds, Kyle McDonell (Submitted on 15 Feb 2021) Comments: Accepted to arXiv. Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI) code： ã¯ã˜ã‚ã« GPT-3ã¯å°‘数サンプル(Few-shot)ã®promptを使ã†äº‹ã§ã€æ§˜ã€…ãªè¨€èªžã‚¿ã‚¹ã‚¯ã«å¯¾ã—ã¦æœ€é«˜æ°´æº–ã®çµæžœã‚’é”æˆ
Browser Dev Tools: The Ultimate Guide 🔥
Contents (Click to expand) â†•ï¸ Design Mode Pretty Print Command Pallet and Super Search Snippets Live Expressions Tracking Changes Console Shorthand Find Unused Code Rendering Panel Network Paint Times Network Timings Inspect Network Requests Performance Identifying Memory Leaks Raw Memory Inspection Test bfcache Full Refresh Lighthouse Page Size Breakdown Record User Flows Advanced User Flow Opera
Bit Twiddling Hacks
By Sean Eron Anderson [email protected] Individually, the code snippets here are in the public domain (unless otherwise noted) — feel free to use them however you please. The aggregate collection and descriptions are © 1997-2005 Sean Eron Anderson. The code and descriptions are distributed in the hope that they will be useful, but WITHOUT ANY WARRANTY and without even the implied warranty of
Stable Diffusionã§morphing - ã™ãŽã‚ƒãƒ¼ã‚“メモ
#stablediffusion 完全ã«ç†è§£ã—㟠pic.twitter.com/IR5yjnL07Y— ã™ãŽã‚ƒãƒ¼ã‚“💯 (@sugyan) August 31, 2022 ã¨ã„ã†ã“ã¨ã§å°‘ã—触ã£ã¦éŠã‚“ã§ã¿ãŸã®ã§ãƒ¡ãƒ¢ã€‚ Stable Diffusion をザックリç†è§£ 先月公開ã•ã‚ŒãŸ Stable Diffusion。 stability.ai 高精度ã§ç¾Žã—ã„ç”»åƒã‚’出力ã§ãる高性能ãªãƒ¢ãƒ‡ãƒ«ãªãŒã‚‰ã€Google Colab ãªã©ã§ã‚‚手軽ã«å‹•ã‹ã›ã‚‹ã—〠Apple silicon ã§ã‚‚ãã‚Œãªã‚Šã«å‹•ã‹ã›ã‚‹ã€ã¨ã„ã†ã®ãŒé…力ã 。 ä¸èº«ã«ã¤ã„ã¦ã¯ 以下ã®è¨˜äº‹ã® "How does Stable Diffusion work?" 以é™ã®ã¨ã“ã‚ãŒåˆ†ã‹ã‚Šã‚„ã™ã„。 huggingface.co 図をãã®ã¾ã¾å¼•ç”¨ã•ã›ã¦ã„ãŸã ã㨠ã¨ã„ã†ä»•çµ„ã¿ã«ãªã£ã¦ã„ã¦ã€å—ã‘å–る入力㯠"User Prompt" 㨠"Late
リリースã€éšœå®³æƒ…å ±ãªã©ã®ã‚µãƒ¼ãƒ“スã®ãŠçŸ¥ã‚‰ã›
最新ã®äººæ°—エントリーã®é…ä¿¡
j次ã®ãƒ–ックマーク
kå‰ã®ãƒ–ックマーク
lã‚ã¨ã§èªã‚€
eコメント一覧を開ã
oページを開ã
GitHub - mrpapercut/nonalpha-js-obfuscator: Non-Alphanumeric JS Obfuscator
Patterns  | Component patterns  | web.dev
{{#tags}}- {{label}}
{{/tags}}