2d:I[923811,["/_next/static/chunks/1ofa4sofdjo4d.js","/_next/static/chunks/3-z86nrqbt-gd.js","/_next/static/chunks/0sjrm73dthvxo.js","/_next/static/chunks/1apkigbabp_jz.js","/_next/static/chunks/1poz9bl8er6mw.js","/_next/static/chunks/1vmlkhm1uomg0.js","/_next/static/chunks/3b4fzrh8g4nh5.js","/_next/static/chunks/3x8n7joqz45xz.js","/_next/static/chunks/1xfxwjxbn4vd9.js","/_next/static/chunks/0717u_rli2lmr.js","/_next/static/chunks/0oq47q7xu8p9i.js","/_next/static/chunks/3rglvh2uekf89.js","/_next/static/chunks/26mjsq_ce17ei.js","/_next/static/chunks/3q8rs7ais2_qk.js","/_next/static/chunks/1nczxl13y2mf0.js","/_next/static/chunks/2egt9jyrzvvz5.js"],"HomeFeed"] 2e:Ta98,

Claude Opus 5.5 마이그레이션 가이드 소개

Anthropic의 Claude Opus 5.5 마이그레이션 가이드는 기존 Messages API 코드를 claude-opus-5-5 로 옮길 때 바꿔야 하는 요청 설정과 응답 처리 코드를 출발 모델별로 정리한 공식 문서입니다. Opus 5.5는 Opus 5에서 문제없이 돌던 요청 중 일부를 400 오류로 거부하기 때문에, 모델 ID만 바꾸면 바로 장애가 날 수 있습니다.

Claude Opus 5.5는 2026년 9월 22일에 공개된 모델로, Anthropic은 이 모델을 오래 실행되는 에이전트 코딩과 지식 노동용으로 소개합니다. 가격은 입력과 출력 100만 토큰당 $4 / $20로 Opus 5의 $5 / $25보다 낮고, 컨텍스트 윈도우는 1M 토큰, 최대 출력은 128K 토큰입니다. 성능과 벤치마크, 안전성 평가는 Claude Opus 5.5 출시 소개 글에서 이미 정리했으므로, 이 글은 기존 코드를 어떻게 옮기는가 에만 집중합니다.

이번 전환이 이전 버전 업그레이드와 다른 점은 모델을 제어하던 손잡이가 줄었다는 것입니다. Opus 5에2f:T7d8,Original post: https://www.reddit.com/r/LocalLLaMA/comments/1woscea/contrastive_language_models/ (sorry I felt it wasn't giving CLM the highlight it deserves) What it is: a new projection head for Qwen3-8B. github: https://github.com/Contrastive-LM/CLM hf: https://huggingface.co/Contrastive-LM At the API and functional interface level, CLM supports everything Jev does—it is not a subset. However, there are important trade-offs in generalization, context scale, and architecture between the two. 1. Functional Parity (Same Primitives) CLM was specifically engineered as an open-weights, self-hostable alternative to TypeSafe AI's Jev . It implements the exact same "System One" decision interface and supports all three of Jev’s core question primitives: Choice : Evaluates a discrete set of candidates and returns a categorical probability distribution. Noul : Outputs a calibrated true/false probability for a proposition or guardrail check. Score : Scores an input against an ordered rubric or scale. Code written for the TypeSafe Jev client can be pointed directly at a clm-serve endpoint with drop-in compatibility (from clm import CLMClient, Choice, Noul, Score). 2. Where CLM Outperforms Jev Latency and Disaggregated Caching: Jev is a proprietary cloud model that evaluates state and question choices jointly. CLM separates the state head from the action head . If an agent has a persistent set of tools or actions, CLM embeds those actions once and caches them. In benchmarks like interactive browser agents and gaming (T-Rex, Super Mario), CLM is 4× to 13× faster than Jev. Open Weights & Fine-Tunability: Jev is a closed API with no user fine-tuning (you can only prompt it via state and question instructions). Because CLM’s heads are tiny open weights (~75 MB), you can fine-tune them on your own agent trajectories. Coding Benchmark Verifiers: When fine-tuned on agent trajectories, CLM achieves state-of-the-art verifier performance on Terminal-Bench 2.1 (87.6%) and DeepSWE (81.6%)30:T912,

GravityOCR 소개

GravityOCR은 문서 이미지를 텍스트, HTML 표, LaTeX 수식으로 옮기는 OCR 모델로, 하나의 가중치가 확산(Diffusion) 방식으로 여러 토큰의 초안을 한 번에 쓰고 같은 모델의 자기회귀(AR) 경로가 그 초안을 검증하는 방식으로 디코딩 속도를 높였습니다. 기반 모델인 GLM-OCR ( GLM-OCR: 0.9B 파라미터로 SOTA 성능을 달성한 초경량 오픈소스 OCR 모델 (feat. Z.ai))과 거의 같은 인식 정확도(OmniDocBench v1.6 Overall 95.16 vs. 95.4831:T7d7,

Google Cloud has become the platform of choice for startups building AI.

Our uniquely complete stack — including a choice of first- and third-party compute and models; our platform for building and managing agents; and our products for securing AI workloads — has emerged as the single most important driver of this growth and it is powering AI development for many of the most exciting and innovative startups in the world.

As a result, startups are choosing to build and run on Google Cloud at a higher rate than they were three years ago, at the start of the AI era.

Given how quickly the technology industry moves in the AI era, the choices startups make can be notable. As we’ve worked together and watch many of th32:Taf2,

pgjev 소개

고객 문의 테이블에서 화가 난 고객이 보낸 문의만 골라내는 작업을 생각해 보면, PostgreSQL 안에서 쓸 수 있는 수단이 생각보다 좁습니다. 전문 검색(Full Text Search)은 키워드가 들어 있는지를 볼 뿐이라 환불을 요구하지는 않았는데 화가 난 문의를 가려내지 못합니다. pgvector로 임베딩(Embedding) 컬럼을 두는 방법은 의미를 다루긴 하지만 돌려받는 값이 유사도 순위라서 조건을 만족하는지 아닌지를 판정해 주지 않고, 임베딩을 만들고 갱신하는 파이프라인을 따로 운영해야 합니다. 남는 선택지는 행을 애플리케이션으로 꺼내 생성형 LLM에 보내고 결과를 파싱해 다시 테이블에 기록하는 방식인데, 이렇게 하면 판정 로직이 SQL 밖으로 나가서 JOIN이나 GROUP BY와 함께 쓸 수 없게 됩니다. 이번에 소개하는 pgjev는 이 판정을 WHERE 절 안에 그대로 두는 PostgreSQL 확장(Extension)입니다.

pgjev가 추가하는 jev(테이블, '조건')은 평범한 boolean 함수입니다. 조건 자리에는 사람이 말하듯 쓴 문장이 들어가고, 테이블의 각 행은 TypeSafe의 Jev 모델이 판정합니다. Jev는 텍스트를 생성하는 대신 보정된 확률(Calibrated Probability)을 돌2b:["$","$L2d",null,{"filterSourceIds":null,"filterFavorites":false,"filterAdminPicks":true,"initialViewMode":"list","initialFeed":{"key":"1|latest||1|0||","rows":[{"id":9071979,"title":"Behind Project Suncatcher, our moonshot to put AI in space","titleKo":"구글, Project Suncatcher로 우주에서 TPU 생존성 검증 프로토타입 발사","hookTitleKo":null,"thumbnailUrl":"https://storage.googleapis.com/gweb-uniblog-publish-prod/images/Project_Suncatcher_social.width-1300.png","url":"https://blog.google/innovation-and-ai/models-and-research/google-research/google-project-suncatcher-facts/","canonicalUrl":"https://blog.google/innovation-and-ai/models-and-research/google-research/google-project-suncatcher-facts","description":"Learn about Project Suncatcher, how we’re testing hardware survival for space, designing cooling systems for AI chips, and more.","score":0,"commentCount":0,"viewCount":0,"bookmarkCount":0,"likeCount":0,"dislikeCount":0,"externalSource":null,"state":"processed","aiCategory":null,"summaryStatus":"done","publishedAt":"2026-09-24T13:00:00+00:00","publishedToSiteAt":"2026-09-24T13:27:34.189086+00:00","createdAt":"2026-09-24T13:25:44.122021+00:00","sourceId":2,"isAdminPick":true,"sourceName":"Google AI Blog","sourceSlug":"google-ai","sourceIconUrl":null,"sourceUrl":"https://blog.google/technology/ai/","sourceTier":1},{"id":9061243,"title":"Claude Opus 5.5 프롬프팅 & 마이그레이션 가이드: Anthropic이 공개한, Opus 5, 4.x, Sonnet 5 코드를 400 오류 없이 옮기는 법","titleKo":"Anthropic, Claude Opus 5.5 마이그레이션 및 프롬프팅 가이드 공개","hookTitleKo":"Anthropic, Claude Opus 5.5 마이그레이션 가이드 공개","thumbnailUrl":"https://discuss.pytorch.kr/uploads/default/optimized/3X/e/a/eadeb9255fbf49cabf62c0171c7edf6f9008e581_2_1024x682.jpeg","url":"https://discuss.pytorch.kr/t/claude-opus-5-5-anthropic-opus-5-4-x-sonnet-5-400/12002","canonicalUrl":"https://discuss.pytorch.kr/t/claude-opus-5-5-anthropic-opus-5-4-x-sonnet-5-400/12002","description":"$2e","score":0,"commentCount":0,"viewCount":0,"bookmarkCount":0,"likeCount":0,"dislikeCount":0,"externalSource":null,"state":"processed","aiCategory":null,"summaryStatus":"done","publishedAt":"2026-09-24T09:30:50+00:00","publishedToSiteAt":"2026-09-24T10:17:10.628712+00:00","createdAt":"2026-09-24T10:14:17.089404+00:00","sourceId":138,"isAdminPick":true,"sourceName":"PyTorchKR 읽을거리","sourceSlug":"pytorchkr-news","sourceIconUrl":null,"sourceUrl":"https://discuss.pytorch.kr/c/news/14","sourceTier":2},{"id":9049073,"title":"JEV almost dead: CLM vs JEV","titleKo":"TypeSafe Jev 대안 오픈소스 CLM 공개, 에이전트 의사결정 지연 4~13배 개선","hookTitleKo":"Jev 대안 CLM 공개, 에이전트 의사결정 4배 빨라","thumbnailUrl":null,"url":"https://www.reddit.com/r/LocalLLaMA/comments/1wouby6/jev_almost_dead_clm_vs_jev/","canonicalUrl":"https://www.reddit.com/r/LocalLLaMA/comments/1wouby6/jev_almost_dead_clm_vs_jev","description":"$2f","score":0,"commentCount":0,"viewCount":0,"bookmarkCount":0,"likeCount":0,"dislikeCount":0,"externalSource":null,"state":"processed","aiCategory":null,"summaryStatus":"done","publishedAt":"2026-09-24T06:40:15+00:00","publishedToSiteAt":"2026-09-24T06:45:30.006891+00:00","createdAt":"2026-09-24T06:45:01.82417+00:00","sourceId":57,"isAdminPick":true,"sourceName":"Reddit","sourceSlug":"reddit","sourceIconUrl":null,"sourceUrl":"https://www.reddit.com","sourceTier":2},{"id":9033844,"title":"GravityOCR: 확산(Diffusion) 및 자기회귀(AR) 기법으로 초안 작성 및 검증을 수행하는 OCR 모델 (feat. Trillion Labs)","titleKo":"Trillion Labs, 확산 및 AR 검증 결합한 초고속 OCR 모델 'GravityOCR' 공개","hookTitleKo":"Trillion Labs, 자기 추측 디코딩 적용한 OCR 모델 GravityOCR 공개","thumbnailUrl":"https://discuss.pytorch.kr/uploads/default/original/3X/2/2/22217a0b4dcaa8ea7ca512069fefd2bed857c262.jpeg","url":"https://discuss.pytorch.kr/t/gravityocr-diffusion-ar-ocr-feat-trillion-labs/11997","canonicalUrl":"https://discuss.pytorch.kr/t/gravityocr-diffusion-ar-ocr-feat-trillion-labs/11997","description":"$30","score":0,"commentCount":0,"viewCount":0,"bookmarkCount":0,"likeCount":0,"dislikeCount":0,"externalSource":null,"state":"processed","aiCategory":null,"summaryStatus":"done","publishedAt":"2026-09-24T02:00:29+00:00","publishedToSiteAt":"2026-09-24T02:09:57.511765+00:00","createdAt":"2026-09-24T02:07:52.970602+00:00","sourceId":138,"isAdminPick":true,"sourceName":"PyTorchKR 읽을거리","sourceSlug":"pytorchkr-news","sourceIconUrl":null,"sourceUrl":"https://discuss.pytorch.kr/c/news/14","sourceTier":2},{"id":9071984,"title":"The three things today's hottest startups are looking for in their AI stack","titleKo":"Google Cloud, 스타트업 AI 스택 구축의 3대 핵심 트렌드 공개","hookTitleKo":null,"thumbnailUrl":"https://storage.googleapis.com/gweb-cloudblog-publish/images/three-things-startups-trends-header.max-2500x2500.png","url":"https://cloud.google.com/blog/topics/startups/the-three-things-todays-hottest-startups-are-looking-for-in-their-ai-stack/","canonicalUrl":"https://cloud.google.com/blog/topics/startups/the-three-things-todays-hottest-startups-are-looking-for-in-their-ai-stack","description":"$31","score":0,"commentCount":0,"viewCount":0,"bookmarkCount":0,"likeCount":0,"dislikeCount":0,"externalSource":null,"state":"processed","aiCategory":null,"summaryStatus":"done","publishedAt":"2026-09-24T00:00:00+00:00","publishedToSiteAt":"2026-09-24T13:28:16.778986+00:00","createdAt":"2026-09-24T13:25:51.860097+00:00","sourceId":133,"isAdminPick":true,"sourceName":"Google Cloud AI","sourceSlug":"google-cloud-ai","sourceIconUrl":null,"sourceUrl":"https://cloud.google.com/blog/products/ai-machine-learning","sourceTier":1},{"id":9016639,"title":"LensVLM-9B by Apple","titleKo":"Apple, 압축된 텍스트 이미지 처리하는 LensVLM-9B 공개","hookTitleKo":null,"thumbnailUrl":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2605.07019/gradient.png","url":"https://huggingface.co/papers/2605.07019","canonicalUrl":"https://huggingface.co/papers/2605.07019","description":"Comments","score":0,"commentCount":0,"viewCount":0,"bookmarkCount":0,"likeCount":0,"dislikeCount":0,"externalSource":null,"state":"processed","aiCategory":null,"summaryStatus":"done","publishedAt":"2026-09-23T20:35:20+00:00","publishedToSiteAt":"2026-09-23T21:01:17.342538+00:00","createdAt":"2026-09-23T21:00:39.663906+00:00","sourceId":11,"isAdminPick":true,"sourceName":"Hacker News","sourceSlug":"hn","sourceIconUrl":"https://news.ycombinator.com/y18.svg","sourceUrl":"https://news.ycombinator.com","sourceTier":2},{"id":9017876,"title":"Gemini 3.8 텍스트 음성 변환","titleKo":"구글, Gemini 3.8 Flash TTS 공개… 자연어 프롬프트로 음성 커스터마이징","hookTitleKo":null,"thumbnailUrl":"https://storage.googleapis.com/gweb-uniblog-publish-prod/images/gemini-audio__keyword__metacard__light.width-1300.png","url":"https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-8-text-to-speech/","canonicalUrl":"https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-8-text-to-speech","description":"

Google이 맞춤형 목소리 설계와 세밀한 연출에 초점을 맞춘 Gemini 3.8 Flash TTS, 대규모 음성 생성의 비용 효율성에 최적화한 Gemini 3.8 Flash-Lite TTS를 출시함
Flash TTS는 자연어로 새로운 목소리를 만들고, 2,000개 이상의 준비된 음성을...
","score":0,"commentCount":0,"viewCount":1,"bookmarkCount":0,"likeCount":0,"dislikeCount":0,"externalSource":null,"state":"processed","aiCategory":null,"summaryStatus":"done","publishedAt":"2026-09-23T19:58:51+00:00","publishedToSiteAt":"2026-09-23T21:29:29.946578+00:00","createdAt":"2026-09-23T21:21:43.267569+00:00","sourceId":2,"isAdminPick":true,"sourceName":"Google AI Blog","sourceSlug":"google-ai","sourceIconUrl":null,"sourceUrl":"https://blog.google/technology/ai/","sourceTier":1},{"id":9004431,"title":"SWE-Bench Pro V2","titleKo":"SWE-Bench Pro V2 공개… 89개 태스크 제거 및 평가 프로토콜 강화","hookTitleKo":null,"thumbnailUrl":"https://labs.scale.com/og?title=SWE-Bench%20Pro%20V2&subtitle=Scale%20Labs&theme=light","url":"https://labs.scale.com/leaderboard/swe_bench_pro_public_v2?utm_source=tldrai","canonicalUrl":"https://labs.scale.com/leaderboard/swe_bench_pro_public_v2","description":"SWE-BENCH PRO V2 releases with 642 tasks from 11 repositories, correcting previous task errors and optimizing evaluation processes. Performance drops for AI models, with OpenAI GPT-5 and Claude Opus 4.1 only scoring around 23% on the public set, highlighting the benchmark's increased challenge and realism compared to SWE-Bench Verified. Notably, top models show consistent results across tasks and languages, while smaller models falter under complex, multi-file scenarios.","score":0,"commentCount":0,"viewCount":0,"bookmarkCount":0,"likeCount":0,"dislikeCount":0,"externalSource":null,"state":"processed","aiCategory":null,"summaryStatus":"done","publishedAt":"2026-09-23T17:15:28.168069+00:00","publishedToSiteAt":"2026-09-23T17:18:10.892703+00:00","createdAt":"2026-09-23T17:15:28.168069+00:00","sourceId":14,"isAdminPick":true,"sourceName":"TLDR AI","sourceSlug":"tldr-ai","sourceIconUrl":"https://tldr.tech/apple-touch-icon.png","sourceUrl":"https://tldr.tech/ai","sourceTier":2},{"id":9003107,"title":"Ringg’s AI agents resolve up to 65% of customer calls with OpenAI","titleKo":"Ringg, OpenAI GPT-5.6 기반 AI 에이전트로 고객 문의 65% 자동 해결","hookTitleKo":null,"thumbnailUrl":"https://images.ctfassets.net/kftzwdyauwt9/773XsnwO9QT1z6x55Ku0BG/678970983193f3c32961714f6eb63320/og.png?w=1600&h=900&fit=fill","url":"https://openai.com/index/ringg","canonicalUrl":"https://openai.com/index/ringg","description":"Using GPT-5.6, Ringg powers multilingual agents across voice, chat, WhatsApp, and web for 90% less cost vs. GPT-4.1.","score":0,"commentCount":0,"viewCount":0,"bookmarkCount":0,"likeCount":0,"dislikeCount":0,"externalSource":null,"state":"processed","aiCategory":null,"summaryStatus":"done","publishedAt":"2026-09-23T16:50:53.968663+00:00","publishedToSiteAt":"2026-09-23T16:52:11.131066+00:00","createdAt":"2026-09-23T16:50:53.970638+00:00","sourceId":1,"isAdminPick":true,"sourceName":"OpenAI Blog","sourceSlug":"openai-blog","sourceIconUrl":null,"sourceUrl":"https://openai.com/blog","sourceTier":1},{"id":9001611,"title":"A new wave of Connected Apps is rolling out to Gemini.","titleKo":"구글, Gemini에 Adobe·Peloton 등 신규 Connected Apps 연동","hookTitleKo":null,"thumbnailUrl":"https://storage.googleapis.com/gweb-uniblog-publish-prod/images/For_blog_16x9.max-1440x810.png","url":"https://blog.google/innovation-and-ai/products/gemini-app/new-connected-apps-gemini/","canonicalUrl":"https://blog.google/innovation-and-ai/products/gemini-app/new-connected-apps-gemini","description":"Gemini is adding new Connected Apps, including Adobe, Airtable, Linear, Peloton and more to help you easily tackle your to-do list.","score":0,"commentCount":0,"viewCount":0,"bookmarkCount":0,"likeCount":0,"dislikeCount":0,"externalSource":null,"state":"processed","aiCategory":null,"summaryStatus":"done","publishedAt":"2026-09-23T16:00:00+00:00","publishedToSiteAt":"2026-09-23T16:27:45.900785+00:00","createdAt":"2026-09-23T16:26:25.130239+00:00","sourceId":2,"isAdminPick":true,"sourceName":"Google AI Blog","sourceSlug":"google-ai","sourceIconUrl":null,"sourceUrl":"https://blog.google/technology/ai/","sourceTier":1},{"id":9001612,"title":"6 new Google Flow Tools built by industry creatives","titleKo":"구글, 크리에이터가 직접 만든 Flow 도구 6종 공개","hookTitleKo":null,"thumbnailUrl":"https://storage.googleapis.com/gweb-uniblog-publish-prod/images/hero_pic.width-2200.format-webp.webp","url":"https://blog.google/innovation-and-ai/models-and-research/google-labs/six-new-tools-built-by-creatives/","canonicalUrl":"https://blog.google/innovation-and-ai/models-and-research/google-labs/six-new-tools-built-by-creatives","description":"Explore 6 new tools in Google Flow built by creatives, for creatives.","score":0,"commentCount":0,"viewCount":0,"bookmarkCount":0,"likeCount":0,"dislikeCount":0,"externalSource":null,"state":"processed","aiCategory":null,"summaryStatus":"done","publishedAt":"2026-09-23T16:00:00+00:00","publishedToSiteAt":"2026-09-23T16:28:00.468975+00:00","createdAt":"2026-09-23T16:26:25.130239+00:00","sourceId":2,"isAdminPick":true,"sourceName":"Google AI Blog","sourceSlug":"google-ai","sourceIconUrl":null,"sourceUrl":"https://blog.google/technology/ai/","sourceTier":1},{"id":8997941,"title":"고객의 취소할 권리를 보장하다","titleKo":"올리브영, '배송 준비 중' 취소 불가 정책 개선… 고객센터 문의 90% 감소","hookTitleKo":null,"thumbnailUrl":"https://oliveyoung.tech/static/dc2d9f45758818e3a159282005f83a42/ac7a9/customer-journey.png","url":"https://oliveyoung.tech/2026-09-23/order-cancellation-shipping-ready/","canonicalUrl":"https://oliveyoung.tech/2026-09-23/order-cancellation-shipping-ready","description":"안녕하세요, 올리브영에서 주문·배송 도메인의 백엔드 개발을 담당하고 있는 데이먼입니다. 오늘은 '배송준비중'이라는 이유로 막혀 있던 주문 취소를, 고객이 앱에서 직접 할 수 있게 바꾼 이야기를 해 보려 합니다. 밤 11시, 한 고객이 결제를 마치고…","score":0,"commentCount":0,"viewCount":0,"bookmarkCount":0,"likeCount":0,"dislikeCount":0,"externalSource":null,"state":"processed","aiCategory":null,"summaryStatus":"done","publishedAt":"2026-09-23T15:21:07.540023+00:00","publishedToSiteAt":"2026-09-23T15:22:31.621786+00:00","createdAt":"2026-09-23T15:21:07.544631+00:00","sourceId":36,"isAdminPick":true,"sourceName":"올리브영","sourceSlug":"oliveyoung-tech","sourceIconUrl":null,"sourceUrl":"https://oliveyoung.tech","sourceTier":1},{"id":8996560,"title":"Two years of OpenAI Academy","titleKo":"OpenAI, 아카데미 2주년 맞아 학습 경로 및 트레이너 프로그램 확장","hookTitleKo":null,"thumbnailUrl":"https://images.ctfassets.net/kftzwdyauwt9/6A4AcxrMdo5oOri7Wm8qNg/34228cf221e3af874539412f71b28a44/two-years-of-openai-academy-seo.png?w=1600&h=900&fit=fill","url":"https://openai.com/index/two-years-of-openai-academy","canonicalUrl":"https://openai.com/index/two-years-of-openai-academy","description":"Marking two years of OpenAI Academy and bringing AI skills to even more communities.","score":0,"commentCount":0,"viewCount":0,"bookmarkCount":0,"likeCount":0,"dislikeCount":0,"externalSource":null,"state":"processed","aiCategory":null,"summaryStatus":"done","publishedAt":"2026-09-23T14:50:44.715875+00:00","publishedToSiteAt":"2026-09-23T14:52:42.309409+00:00","createdAt":"2026-09-23T14:50:44.718067+00:00","sourceId":1,"isAdminPick":true,"sourceName":"OpenAI Blog","sourceSlug":"openai-blog","sourceIconUrl":null,"sourceUrl":"https://openai.com/blog","sourceTier":1},{"id":8990597,"title":"Developers want more efficient software. Here’s what over 1000 GitHub users told us they need.","titleKo":"GitHub-Yale 설문, 개발자 80%는 에너지 효율 코드 도구 원하지만 행동은 미미","hookTitleKo":null,"thumbnailUrl":"https://github.blog/wp-content/uploads/2026/09/Screenshot-2026-09-22-at-6.34.24-PM.png?w=1536","url":"https://github.blog/news-insights/research/developers-want-more-efficient-software-heres-what-over-1000-github-users-told-us-they-need/","canonicalUrl":"https://github.blog/news-insights/research/developers-want-more-efficient-software-heres-what-over-1000-github-users-told-us-they-need","description":"
New research from GitHub and Yale Program on Climate Change Communication finds strong demand for tools, measurement, and practical guidance that can help developers reduce wasted compute.
\n
The post Developers want more efficient software. Here’s what over 1000 GitHub users told us they need. appeared first on The GitHub Blog.
","score":0,"commentCount":0,"viewCount":0,"bookmarkCount":0,"likeCount":0,"dislikeCount":0,"externalSource":null,"state":"processed","aiCategory":null,"summaryStatus":"done","publishedAt":"2026-09-23T13:00:00+00:00","publishedToSiteAt":"2026-09-23T13:14:29.523642+00:00","createdAt":"2026-09-23T13:12:39.076791+00:00","sourceId":53,"isAdminPick":true,"sourceName":"GitHub Blog","sourceSlug":"github-blog","sourceIconUrl":null,"sourceUrl":"https://github.blog","sourceTier":1},{"id":8994064,"title":"Show HN: Jevper – the Jev interface on top of any OpenAI-compatible model","titleKo":"OpenAI 호환 모델용 Jev 분류 래퍼 'Jevper' 공개, 자동 로그프롭스 감지 기능 탑재","hookTitleKo":null,"thumbnailUrl":"https://opengraph.githubassets.com/8028c80b4b1200b1c39789820632824cf32fe652501f9caceba2bdc855a2f8f9/zhulinchng/jevper","url":"https://github.com/zhulinchng/jevper","canonicalUrl":"https://github.com/zhulinchng/jevper","description":"Comments","score":0,"commentCount":0,"viewCount":0,"bookmarkCount":0,"likeCount":0,"dislikeCount":0,"externalSource":null,"state":"processed","aiCategory":null,"summaryStatus":"done","publishedAt":"2026-09-23T12:25:21+00:00","publishedToSiteAt":"2026-09-23T14:14:37.423982+00:00","createdAt":"2026-09-23T14:13:16.611825+00:00","sourceId":11,"isAdminPick":true,"sourceName":"Hacker News","sourceSlug":"hn","sourceIconUrl":"https://news.ycombinator.com/y18.svg","sourceUrl":"https://news.ycombinator.com","sourceTier":2},{"id":8968191,"title":"연 300시간을 아낀 AI 상담 서비스","titleKo":"토스뱅크, AI 상담 도입으로 연 300시간 절감","hookTitleKo":null,"thumbnailUrl":"https://static.toss.im/ipd-tcs/toss_core/live/872256fb-c21a-4e4c-8124-f61bb1e16d36/image.png","url":"https://toss.tech/article/AI_chatbot","canonicalUrl":"https://toss.tech/article/AI_chatbot","description":"전화 문의를 줄이려다 사용자의 경험에서 답을 찾은 이야기예요.","score":0,"commentCount":0,"viewCount":0,"bookmarkCount":0,"likeCount":0,"dislikeCount":0,"externalSource":null,"state":"processed","aiCategory":null,"summaryStatus":"done","publishedAt":"2026-09-23T06:39:00+00:00","publishedToSiteAt":"2026-09-23T06:52:10.210629+00:00","createdAt":"2026-09-23T06:51:35.59456+00:00","sourceId":24,"isAdminPick":true,"sourceName":"토스","sourceSlug":"toss-tech","sourceIconUrl":"https://static.toss.im/tds/favicon/apple-touch-icon-152x152.png","sourceUrl":"https://toss.tech","sourceTier":1},{"id":8962677,"title":"개인화 추천을 위한 랭킹 모델 개발기","titleKo":"카카오, GDCN과 MoE 기반 개인화 랭킹 모델로 유효재생률 20% 개선","hookTitleKo":null,"thumbnailUrl":"https://t1.kakaocdn.net/kakao_tech/media/cbd8729101a000001.png","url":"https://tech.kakao.com/posts/837","canonicalUrl":"https://tech.kakao.com/posts/837","description":"안녕하세요! 카카오에서 유저 편의성을 높이는 데 관심이 많은 토리입니다.\n카카오에는 정말 다양한 서비스와 그 안에 쌓이는 방대한 데이터가 있습니다. 이 많은 콘텐츠와 유저를 잘 이어주는 역할을 하는 것이 바로 추천이고, 그중에서도 \"얼마나 이 유저에게 맞는 콘텐츠를 보여주는가\"는 서비스 체류시간과 만족도에 직결되는 핵심 문제입니다. 그래서 세그먼트 단위로 동작하던 추천 로직을, 유저 개인의 히스토리를 반영하는 개인화 랭킹 모델로 교체하는 프로젝트를 진행...","score":0,"commentCount":0,"viewCount":0,"bookmarkCount":0,"likeCount":0,"dislikeCount":0,"externalSource":null,"state":"processed","aiCategory":null,"summaryStatus":"done","publishedAt":"2026-09-23T01:01:00+00:00","publishedToSiteAt":"2026-09-23T05:23:52.414998+00:00","createdAt":"2026-09-23T05:21:59.940714+00:00","sourceId":23,"isAdminPick":true,"sourceName":"카카오","sourceSlug":"kakao-tech","sourceIconUrl":null,"sourceUrl":"https://tech.kakao.com/posts","sourceTier":1},{"id":8947887,"title":"Drop - gVisor를 지원하는 루트 권한 없는 Linux 샌드박스","titleKo":"AI 에이전트 및 서드파티 프로그램 격리를 위한 루트 권한 없는 Linux 샌드박스 'Drop' 공개","hookTitleKo":null,"thumbnailUrl":"https://droprun.sh/og-image.png","url":"https://droprun.sh/","canonicalUrl":"https://droprun.sh/","description":"
- Drop은 익숙한 작업 환경을 유지하면서 프로그램과 코딩 에이전트를 격리하는 Linux 샌드박스로, 기존 배포판과 설치된 프로그램을 그대로 사용함
- 환경마다 별도의 홈 디렉터리를 제공하고 원래 홈은 숨김 처리해, 에이전트의 잘못된 삭제 명령이나 비밀 정보 접근...
  ","score":0,"commentCount":0,"viewCount":0,"bookmarkCount":0,"likeCount":0,"dislikeCount":0,"externalSource":null,"state":"processed","aiCategory":null,"summaryStatus":"done","publishedAt":"2026-09-23T00:37:52+00:00","publishedToSiteAt":"2026-09-23T01:12:31.152649+00:00","createdAt":"2026-09-23T01:10:54.921451+00:00","sourceId":12,"isAdminPick":true,"sourceName":"GeekNews","sourceSlug":"geeknews","sourceIconUrl":"https://news.hada.io/apple-touch-icon.png","sourceUrl":"https://news.hada.io","sourceTier":2},{"id":8947889,"title":"Husky - MLX보다 최대 4.5배 빠른 모델 특화 추론 엔진","titleKo":"Underdog, MLX보다 최대 4.5배 빠른 모델 특화 추론 엔진 'Husky' 공개","hookTitleKo":null,"thumbnailUrl":"https://husky.underdog.ai/husky-card.jpg","url":"https://husky.underdog.ai/","canonicalUrl":"https://husky.underdog.ai/","description":"
  - 여러 모델을 지원하는 대신 4B 모델 Woof 하나에 최적화한 애플 실리콘 추론 엔진으로, 메일/일정 관리/회의록 요약을 기기에서 처리하는 맥용 개인 AI 비서 앱 Underdog에 적용됨
  - M5 Max 비교에서 같은 가중치와 동일한 출력으로 16개 작업 모두 MLX보다 ...
    ","score":0,"commentCount":0,"viewCount":0,"bookmarkCount":0,"likeCount":0,"dislikeCount":0,"externalSource":null,"state":"processed","aiCategory":null,"summaryStatus":"done","publishedAt":"2026-09-23T00:22:34+00:00","publishedToSiteAt":"2026-09-23T01:15:47.676921+00:00","createdAt":"2026-09-23T01:10:54.921451+00:00","sourceId":12,"isAdminPick":true,"sourceName":"GeekNews","sourceSlug":"geeknews","sourceIconUrl":"https://news.hada.io/apple-touch-icon.png","sourceUrl":"https://news.hada.io","sourceTier":2},{"id":8983796,"title":"Grab and OpenAI bring practical AI skills to Southeast Asia","titleKo":"Grab·OpenAI, 동남아 파트너 3만 명 대상 AI 스킬 교육 프로그램 발표","hookTitleKo":null,"thumbnailUrl":"https://images.ctfassets.net/kftzwdyauwt9/4xEpNOGJNAtSetkblkQrHz/aed1d68ba4955df9170a42d60a9d43b8/Grab-and-OpenAI-today-announced-_GO-Forward-with-AI-scaled.jpg?w=3840&q=90&fm=webp","url":"https://openai.com/index/grab-openai-ai-skills-southeast-asia","canonicalUrl":"https://openai.com/index/grab-openai-ai-skills-southeast-asia","description":"OpenAI and Grab launch GO Forward with AI, a regional programme helping 30,000 partners build practical AI skills across Southeast Asia.","score":0,"commentCount":0,"viewCount":0,"bookmarkCount":0,"likeCount":0,"dislikeCount":0,"externalSource":null,"state":"processed","aiCategory":null,"summaryStatus":"done","publishedAt":"2026-09-23T00:00:00+00:00","publishedToSiteAt":"2026-09-23T11:14:33.089029+00:00","createdAt":"2026-09-23T11:12:39.972676+00:00","sourceId":1,"isAdminPick":true,"sourceName":"OpenAI Blog","sourceSlug":"openai-blog","sourceIconUrl":null,"sourceUrl":"https://openai.com/blog","sourceTier":1},{"id":8943252,"title":"pgjev: 임베딩이나 벡터 컬럼 없이 WHERE 절에서 자연어로 행을 필터링하는 PostgreSQL 확장","titleKo":"PostgreSQL 확장 'pgjev' 공개: 임베딩 없이 SQL WHERE 절에서 자연어 필터링 지원","hookTitleKo":"PostgreSQL 확장 pgjev 공개, 임베딩 없이 자연어 필터링 지원","thumbnailUrl":"https://discuss.pytorch.kr/uploads/default/optimized/3X/f/2/f215d54cdbcd05b687f8ea97273ec6d4652e3154_2_2056x844.png","url":"https://discuss.pytorch.kr/t/pgjev-where-postgresql/11984","canonicalUrl":"https://discuss.pytorch.kr/t/pgjev-where-postgresql/11984","description":"$32","score":0,"commentCount":0,"viewCount":0,"bookmarkCount":0,"likeCount":0,"dislikeCount":0,"externalSource":null,"state":"processed","aiCategory":null,"summaryStatus":"done","publishedAt":"2026-09-22T22:00:37+00:00","publishedToSiteAt":"2026-09-22T23:47:21.843294+00:00","createdAt":"2026-09-22T23:46:19.534419+00:00","sourceId":138,"isAdminPick":true,"sourceName":"PyTorchKR 읽을거리","sourceSlug":"pytorchkr-news","sourceIconUrl":null,"sourceUrl":"https://discuss.pytorch.kr/c/news/14","sourceTier":2},{"id":8932536,"title":"Better prompt caching for GPT-6","titleKo":"OpenAI, GPT-6 모델용 프롬프트 캐싱 최적화 가이드 및 진단 도구 공개","hookTitleKo":null,"thumbnailUrl":"https://images.ctfassets.net/kftzwdyauwt9/29DVnNtIz2bkMRa1wWULXs/062723af9185edeb09c27071e68b589d/image__17_.png?w=3840&q=90&fm=webp","url":"https://openai.com/index/better-prompt-caching-for-gpt-6","canonicalUrl":"https://openai.com/index/better-prompt-caching-for-gpt-6","description":"Learn how GPT-6 improves prompt caching with higher cache hit rates, new diagnostics, explicit breakpoints, and controls that reduce latency and costs.","score":0,"commentCount":0,"viewCount":0,"bookmarkCount":0,"likeCount":0,"dislikeCount":0,"externalSource":null,"state":"processed","aiCategory":null,"summaryStatus":"done","publishedAt":"2026-09-22T20:44:08.821238+00:00","publishedToSiteAt":"2026-09-22T20:45:40.648928+00:00","createdAt":"2026-09-22T20:44:08.823421+00:00","sourceId":1,"isAdminPick":true,"sourceName":"OpenAI Blog","sourceSlug":"openai-blog","sourceIconUrl":null,"sourceUrl":"https://openai.com/blog","sourceTier":1},{"id":9004442,"title":"Meta admits Muse's likeness to OpenClaw isn't a coincidence","titleKo":"Meta, AI 어시스턴트 'Muse'가 OpenClaw에서 큰 영감 받았다고 인정","hookTitleKo":null,"thumbnailUrl":"https://techcrunch.com/wp-content/uploads/2026/09/meta-muse-app-GettyImages-2293792137.jpg?w=1024","url":"https://techcrunch.com/2026/09/22/meta-admits-muses-likeness-to-openclaw-isnt-a-coincidence/?utm_source=tldrai","canonicalUrl":"https://techcrunch.com/2026/09/22/meta-admits-muses-likeness-to-openclaw-isnt-a-coincidence","description":"Meta admits Muse draws heavy inspiration from the open-source project OpenClaw, despite being built from scratch.","score":0,"commentCount":0,"viewCount":0,"bookmarkCount":0,"likeCount":0,"dislikeCount":0,"externalSource":null,"state":"processed","aiCategory":null,"summaryStatus":"done","publishedAt":"2026-09-22T19:09:11+00:00","publishedToSiteAt":"2026-09-23T17:20:34.318638+00:00","createdAt":"2026-09-23T17:15:28.168069+00:00","sourceId":14,"isAdminPick":true,"sourceName":"TLDR AI","sourceSlug":"tldr-ai","sourceIconUrl":"https://tldr.tech/apple-touch-icon.png","sourceUrl":"https://tldr.tech/ai","sourceTier":2},{"id":8928544,"title":"Introducing GPT-6 Sol and Luna","titleKo":"OpenAI, GPT-6 Sol 및 Luna 출시… GPT-5.6 대비 API 가격 50% 인하","hookTitleKo":null,"thumbnailUrl":"https://images.ctfassets.net/kftzwdyauwt9/7fvmMcBIvMTXxVJh7NEMt1/48a75e624699bb1e218366c1d3522048/gpt-6-sol-luna-seo.png?w=1600&h=900&fit=fill","url":"https://openai.com/index/introducing-gpt-6-sol-and-luna","canonicalUrl":"https://openai.com/index/introducing-gpt-6-sol-and-luna","description":"Meet GPT-6 Sol and Luna, two models that bring frontier intelligence to everyday work with different balances of capability and cost.","score":0,"commentCount":0,"viewCount":0,"bookmarkCount":0,"likeCount":0,"dislikeCount":0,"externalSource":null,"state":"processed","aiCategory":null,"summaryStatus":"done","publishedAt":"2026-09-22T18:00:00+00:00","publishedToSiteAt":"2026-09-22T19:16:25.102427+00:00","createdAt":"2026-09-22T19:14:57.957043+00:00","sourceId":1,"isAdminPick":true,"sourceName":"OpenAI Blog","sourceSlug":"openai-blog","sourceIconUrl":null,"sourceUrl":"https://openai.com/blog","sourceTier":1}],"total":293}}]

최신 AI 뉴스

추천구글, Project Suncatcher로 우주에서 TPU 생존성 검증 프로토타입 발사

추천Anthropic, Claude Opus 5.5 마이그레이션 가이드 공개

추천Jev 대안 CLM 공개, 에이전트 의사결정 4배 빨라

추천Trillion Labs, 자기 추측 디코딩 적용한 OCR 모델 GravityOCR 공개

추천Google Cloud, 스타트업 AI 스택 구축의 3대 핵심 트렌드 공개

추천Apple, 압축된 텍스트 이미지 처리하는 LensVLM-9B 공개

추천구글, Gemini 3.8 Flash TTS 공개… 자연어 프롬프트로 음성 커스터마이징

추천SWE-Bench Pro V2 공개… 89개 태스크 제거 및 평가 프로토콜 강화

추천Ringg, OpenAI GPT-5.6 기반 AI 에이전트로 고객 문의 65% 자동 해결

추천구글, Gemini에 Adobe·Peloton 등 신규 Connected Apps 연동

추천구글, 크리에이터가 직접 만든 Flow 도구 6종 공개

추천올리브영, '배송 준비 중' 취소 불가 정책 개선… 고객센터 문의 90% 감소

추천OpenAI, 아카데미 2주년 맞아 학습 경로 및 트레이너 프로그램 확장

추천GitHub-Yale 설문, 개발자 80%는 에너지 효율 코드 도구 원하지만 행동은 미미

추천OpenAI 호환 모델용 Jev 분류 래퍼 'Jevper' 공개, 자동 로그프롭스 감지 기능 탑재

추천토스뱅크, AI 상담 도입으로 연 300시간 절감

추천카카오, GDCN과 MoE 기반 개인화 랭킹 모델로 유효재생률 20% 개선

추천AI 에이전트 및 서드파티 프로그램 격리를 위한 루트 권한 없는 Linux 샌드박스 'Drop' 공개

추천Underdog, MLX보다 최대 4.5배 빠른 모델 특화 추론 엔진 'Husky' 공개

추천Grab·OpenAI, 동남아 파트너 3만 명 대상 AI 스킬 교육 프로그램 발표

추천PostgreSQL 확장 pgjev 공개, 임베딩 없이 자연어 필터링 지원

추천OpenAI, GPT-6 모델용 프롬프트 캐싱 최적화 가이드 및 진단 도구 공개

추천Meta, AI 어시스턴트 'Muse'가 OpenClaw에서 큰 영감 받았다고 인정

추천OpenAI, GPT-6 Sol 및 Luna 출시… GPT-5.6 대비 API 가격 50% 인하