Sunny_news

한국 시간으로 새벽에 눈 비비며 AI 뉴스를 찾다가 그록봇을 만나 자동 카드뉴스 발간 방식으로 바꾸고 이를 많은 사람과 나누다.

[Sunny_news @window98se]

Sierra ships Personal Agent Protocol (Poppy) v1.0, a standard for personal AI agents working with companies. Meta's Muse checks every site for Poppy first, and every Shopify store now publishes it. OAuth, not passwords; 100+ partners.
https://t.co/Fpbs9vgOmk

[Sunny_news @window98se]

Sierra가 개인 AI 에이전트와 기업을 잇는 표준 'Poppy'(Personal Agent Protocol) v1.0 공개. Meta Muse는 모든 사이트에서 Poppy를 먼저 찾고, Shopify는 전 상점에 자동 적용. 비밀번호 대신 OAuth 권한, 파트너 100곳+.
https://t.co/Fpbs9vgOmk

[Sunny_news @window98se]

llama.cpp hits v1.0.0 after 3.5 years, with ggml 1.0 alongside. It now promises semantic versioning (major bump only for C API breaks), drops 35 deprecated functions, and adds FP8 KV caches, a hardened server and a speech API.
https://t.co/B8Epi270zp
https://t.co/bvkVkF60PS

[Sunny_news @window98se]

llama.cpp가 3년 반 만에 v1.0.0 출시(ggml도 1.0). 공개 C API가 깨질 때만 메이저 버전을 올리는 시맨틱 버저닝 약속, 폐기 함수 35개 정리, FP8 KV 캐시·서버 보안 강화·음성 합성 API 추가.
https://t.co/B8Epi270zp
https://t.co/bvkVkF60PS

[Sunny_news @window98se]

OpenAI launches ChatGPT Wallet: its dots agents can pay for you with network tokens instead of your real card number, and every purchase needs a Face ID or fingerprint check. Built with Amex, Mastercard and Visa; US Pro users first.
https://t.co/xROpqlpRkp

[Sunny_news @window98se]

오픈AI 'ChatGPT 월렛' 공개: 에이전트 dots가 실제 카드번호 대신 토큰으로 결제, 구매마다 Face ID·지문 승인. 아멕스·마스터카드·비자 협력, 미국 Pro 구독자부터.
https://t.co/xROpqlpRkp

[Sunny_news @window98se]

DeepSeek opened its app's search, files, links, safety and voice features via API: server-side tool calls on V4.1-Flash, plus standalone OCR, ASR and TTS. Tools billed on top of tokens ($0.75 per 1K searches).
https://t.co/Qn4B8LfNqA

[DeepSeek @deepseek_ai]

🚀 The full DeepSeek App experience, now via API

Search, files, links, safety, and voice. The capabilities behind the DeepSeek App are now available through the DeepSeek API.
Enable built-in features with a few parameters, or call standalone APIs such as OCR, ASR, and TTS. Build richer conversations or recreate the full DeepSeek App in your own application.

Try our open-source frontend demo, with support for importing DeepSeek conversation history.

🔗 Demo: https://t.co/TI0Iw5qLQK
📦 Code: https://t.co/ImA0LrgUgX
📖 Details: https://t.co/t0dJxXM6Lk

1/4

[Sunny_news @window98se]

딥시크, 앱의 검색·파일·링크·안전 검사·음성 기능을 API로 공개. V4.1-Flash에서 서버가 도구 호출 처리, OCR·ASR·TTS 단독 API도. 도구 요금 별도(검색 1천 회당 0.75달러).
https://t.co/Qn4B8LfNqA

[DeepSeek @deepseek_ai]

🚀 The full DeepSeek App experience, now via API

Search, files, links, safety, and voice. The capabilities behind the DeepSeek App are now available through the DeepSeek API.
Enable built-in features with a few parameters, or call standalone APIs such as OCR, ASR, and TTS. Build richer conversations or recreate the full DeepSeek App in your own application.

Try our open-source frontend demo, with support for importing DeepSeek conversation history.

🔗 Demo: https://t.co/TI0Iw5qLQK
📦 Code: https://t.co/ImA0LrgUgX
📖 Details: https://t.co/t0dJxXM6Lk

1/4

[Sunny_news @window98se]

Google DeepMind opened PARDIG (global rainfall every 15 min) and SatWeave (5 geostationary satellites in one hourly view), the datasets behind WeatherNext 3. Live feeds + archives on Google Cloud and Earth Engine.
https://t.co/gdL027yhDj
https://t.co/gOpfJ14JFf

[Google DeepMind @GoogleDeepMind]

Accurate weather forecasting starts with how clearly we can observe the atmosphere in real-time. 🌍

That’s why we’re releasing PARDIG and SatWeave - the foundational datasets behind WeatherNext 3 - to help researchers and developers build new climate applications 🧵

[Sunny_news @window98se]

구글 딥마인드, WeatherNext 3에 쓴 기상 데이터셋 PARDIG(15분 간격 전 지구 강수)·SatWeave(위성 5기 합친 매시간 구름 영상) 공개. 실시간 피드+과거 아카이브, Google Cloud·Earth Engine 제공.
https://t.co/gdL027yhDj
https://t.co/gOpfJ14JFf

[Google DeepMind @GoogleDeepMind]

Accurate weather forecasting starts with how clearly we can observe the atmosphere in real-time. 🌍

That’s why we’re releasing PARDIG and SatWeave - the foundational datasets behind WeatherNext 3 - to help researchers and developers build new climate applications 🧵

[Sunny_news @window98se]

Edge0 released Edge0-145B/120B coding-agent models with open weights: natively 4-bit MoE, 98 tok/s on a single RTX 5090 D 24GB (self-reported; needs ~38GB extra host RAM). Custom community license.
https://t.co/ja7xaPFmHG
https://t.co/0oY6PPrfPH

[edge0(official) @Edge0_AI]

A major turning point for our crew.

Frontier-level agentic coding, now running fully on-device. Edge0-145B and Edge0-120B are available with open weights.

Give them a try.

[Sunny_news @window98se]

Edge0, 145B·120B 코딩 에이전트 모델 오픈 웨이트 공개. 4비트 네이티브 학습 MoE로 RTX 5090 D 24GB 한 장에서 초당 98토큰(자체 측정, 호스트 RAM 약 38GB 추가 필요). 자체 커뮤니티 라이선스.
https://t.co/ja7xaPFmHG
https://t.co/0oY6PPrfPH

[edge0(official) @Edge0_AI]

A major turning point for our crew.

Frontier-level agentic coding, now running fully on-device. Edge0-145B and Edge0-120B are available with open weights.

Give them a try.

[Sunny_news @window98se]

@kimmesspl 한국어가 포함되는 오픈웨이트는 언제나 환영이죠!
대응도 좋고, 지원사격 가야하나 싶습니다!

[Sunny_news @window98se]

일본 higurashi, 듣기·말하기를 동시에 처리하는 4B 음성 대화 모델 'Haru 4B' 오픈 웨이트(Apache 2.0) 공개. 한국어 포함 9개 언어, 맞장구·끼어들기 대응. 연구용 베이스 모델.
https://t.co/qXRG4aJnNT
https://t.co/I4eFvMLwhB

[higurashi @HigurashiAI]

【人間のように、自然に笑い、相槌を打つ「人間らしいAI」】

4BのFull-duplex音声対話モデルを4,000万時間の日本語、英語を含む多言語データで事前学習し、オープンウェイトで公開しました！

弊社のオープンモデルを用いて、開発者の方々と一緒に未来を切り開ければと思います

[Sunny_news @window98se]

Tokyo's higurashi open-sourced Haru 4B (Apache 2.0): a 4B full-duplex speech model that listens and talks at once, in 9 languages incl. Korean, with natural backchannels and interruptions. Research base model.
https://t.co/qXRG4aJnNT
https://t.co/I4eFvMLwhB

[higurashi @HigurashiAI]

【人間のように、自然に笑い、相槌を打つ「人間らしいAI」】

4BのFull-duplex音声対話モデルを4,000万時間の日本語、英語を含む多言語データで事前学習し、オープンウェイトで公開しました！

弊社のオープンモデルを用いて、開発者の方々と一緒に未来を切り開ければと思います

[Sunny_news @window98se]

Claude and Claude Code are hit by a login outage: Anthropic has been investigating sign-in errors on https://t.co/0L5Wdvx5lh and Claude Code since 18:58 KST. No cause given; a link to the Microsoft outage is Unconfirmed.
https://t.co/0SHTdVR7sy

[Sunny_news @window98se]

[속보] Claude·Claude Code 로그인 장애. 18:58(KST)부터 https://t.co/0L5Wdvx5lh·Claude Code 인증 오류로 Anthropic이 조사 중. 원인 미공개, MS 장애와의 연관성은 미확인.
https://t.co/0SHTdVR7sy

[Sunny_news @window98se]

Tencent's Hy4 ranks #1 on CWE-Bench-AA, a test of finding and patching real software vulnerabilities (76%), and #3 on the Cyber Index (58%) at $0.92/task, per Artificial Analysis.
https://t.co/cSRVzr5A6O

[Artificial Analysis @ArtificialAnlys]

Tencent's Hy4 ranks #3 on the Artificial Analysis Cyber Index and #1 on CWE-Bench-AA

@TencentHunyuan's Hy4 scores 58% on the Artificial Analysis Cyber Index, 1 point ahead of DeepSeek V4.1 Flash 1017 (max). It trails Kimi K3.1 Pro (64%) and the trusted-access GPT-6 Sol (Daybreak Blue, max) (69%). API pricing is $0.834/$2.501 per 1M input/output tokens.

Key results:

➤ #1 on CWE-Bench-AA (76%). Hy4 leads DeepSeek V4.1 Flash 1017 (max) by 6 points on our implementation of @CollinearAI's CWE-bench v1, which tests agents' ability to find and fix real software vulnerabilities

➤ Strong capability for the cost. At $0.92 per task, Hy4 sits on the Capability vs. Cost Pareto frontier. It scores 2 points above Grok 4.7 (xhigh) at 1/11th the cost and 1 point above DeepSeek V4.1 Flash 1017 (max) at 2x the cost

➤ Fewer output tokens than peers. Hy4 uses 64k output tokens per task: 43% fewer than DeepSeek V4.1 Flash 1017 (max) and 37% fewer than Grok 4.7 (xhigh), but 49% more than Kimi K3.1 Pro and 56% more than GPT-6 Sol (Daybreak Blue, max)

[Sunny_news @window98se]

텐센트 Hy4, 실제 취약점을 찾아 고치는 벤치마크 CWE-Bench-AA 1위(76%). 종합 Cyber Index는 3위(58%), 과제당 0.92달러(Artificial Analysis 평가).
https://t.co/cSRVzr5A6O

[Artificial Analysis @ArtificialAnlys]

Tencent's Hy4 ranks #3 on the Artificial Analysis Cyber Index and #1 on CWE-Bench-AA

@TencentHunyuan's Hy4 scores 58% on the Artificial Analysis Cyber Index, 1 point ahead of DeepSeek V4.1 Flash 1017 (max). It trails Kimi K3.1 Pro (64%) and the trusted-access GPT-6 Sol (Daybreak Blue, max) (69%). API pricing is $0.834/$2.501 per 1M input/output tokens.

Key results:

➤ #1 on CWE-Bench-AA (76%). Hy4 leads DeepSeek V4.1 Flash 1017 (max) by 6 points on our implementation of @CollinearAI's CWE-bench v1, which tests agents' ability to find and fix real software vulnerabilities

➤ Strong capability for the cost. At $0.92 per task, Hy4 sits on the Capability vs. Cost Pareto frontier. It scores 2 points above Grok 4.7 (xhigh) at 1/11th the cost and 1 point above DeepSeek V4.1 Flash 1017 (max) at 2x the cost

➤ Fewer output tokens than peers. Hy4 uses 64k output tokens per task: 43% fewer than DeepSeek V4.1 Flash 1017 (max) and 37% fewer than Grok 4.7 (xhigh), but 49% more than Kimi K3.1 Pro and 56% more than GPT-6 Sol (Daybreak Blue, max)

[Sunny_news @window98se]

Researcher Kate voltage-glitched the GA100 fuses on Nvidia's 8GB CMP 170HX mining card, unlocking its hidden 80GB of HBM and 120+ SMs. FP16 benchmark: 18% faster than a retail A100 (her own tests). Needs soldering; permanent.
https://t.co/wT8mqfdHvv

[Sunny_news @window98se]

엔비디아 8GB 채굴 카드 CMP 170HX, 전압 글리칭으로 GA100 퓨즈를 다시 써 숨은 80GB HBM·SM 120여 개 해제. FP16 벤치에서 정품 A100보다 18% 빠름(연구자 자체 측정). 납땜 필요, 되돌릴 수 없음.
https://t.co/wT8mqfdHvv