https://24-ai.news/hr/news/2026-07-17/amazon-quick-agentic-sales-teammate/ 24 AI hr 2026-07-17T00:00:00Z AWS: Amazon Quick je agentski AI kolega za prodajne timove koji automatizira CRM i pripremu sastanaka https://24-ai.news/hr/news/2026-07-17/anthropic-claude-code-v2-1-212/ 24 AI hr 2026-07-17T00:00:00Z Anthropic: Claude Code v2.1.212 donosi redizajnirani /fork, novi /resume picker i sigurnosne popravke https://24-ai.news/hr/news/2026-07-17/arxiv-llm-gpu-kernels-atrex/ 24 AI hr 2026-07-17T00:00:00Z arXiv:2607.14541: Atrex-Bench otkriva da LLM-generirani GPU kerneli dosežu tek 10% ciljane hardverske brzine https://24-ai.news/hr/news/2026-07-17/arxiv-mcpevol-bench-agent-tools/ 24 AI hr 2026-07-17T00:00:00Z arXiv:2607.14642: MCPEvol-Bench pokazuje da i GPT-5.4 i Claude gube preciznost kad se mijenjaju MCP alati https://24-ai.news/hr/news/2026-07-17/arxiv-physical-danger-prism-safety/ 24 AI hr 2026-07-17T00:00:00Z arXiv:2607.15218: Kad su riječi sigurne, ali radnje ubijaju — PRISM razdvaja fizičku opasnost od teksta https://24-ai.news/hr/news/2026-07-17/arxiv-pretraining-data-poisoning/ 24 AI hr 2026-07-17T00:00:00Z arXiv:2607.15267: Trovanje podataka za pretreniranje jezičnih modela kroz računalnu propagandu na forumima https://24-ai.news/hr/news/2026-07-17/arxiv-transcoders-deception/ 24 AI hr 2026-07-17T00:00:00Z arXiv:2607.14791: Transkoderi otkrivaju kako laž nastaje unutar internih sklopova jezičnog modela Qwen3-4B https://24-ai.news/hr/news/2026-07-17/github-copilot-review-mobile-updates/ 24 AI hr 2026-07-17T00:00:00Z GitHub: Copilot code review dobiva firewall i custom runnere, mobitel dobiva cloud agent popravak komentara https://24-ai.news/hr/news/2026-07-17/github-cost-of-saying-yes/ 24 AI hr 2026-07-17T00:00:00Z GitHub: cijena pisanja koda je pala, cijena vlasništva nije — novi editorijal o ekonomiji inženjeringa https://24-ai.news/hr/news/2026-07-17/nvidia-vera-rubin-post-training/ 24 AI hr 2026-07-17T00:00:00Z NVIDIA: Vera Rubin platforma trenira najveće AI modele s četvrtinom GPU-a Blackwell generacije zahvaljujući codesignu https://24-ai.news/hr/news/2026-07-17/openai-ai-scorecard-roi-metrics/ 24 AI hr 2026-07-17T00:00:00Z OpenAI: CFO Sarah Friar predlaže AI scorecard sa četiri metrike za mjerenje ROI-ja AI ulaganja tvrtki https://24-ai.news/hr/news/2026-07-17/smartsheet-remote-mcp-server-aws/ 24 AI hr 2026-07-17T00:00:00Z AWS: Smartsheet uštedio 3 milijarde tokena izgradnjom remote MCP servera s Bedrock i Neptune knowledge grafom https://24-ai.news/hr/news/2026-07-17/vllm-ci-benchmarking-quality/ 24 AI hr 2026-07-17T00:00:00Z vLLM: kako 266 CI jobova i noćni benchmarking čuvaju produkcijsku kvalitetu open-source LLM enginea za serving https://24-ai.news/hr/news/2026-07-17/xai-grok-4-5-api-release/ 24 AI hr 2026-07-17T00:00:00Z xAI: Grok 4.5 dostupan na API-ju s podesivim reasoning-effort razinama i cijenom od 2 dolara po tokenu https://24-ai.news/en/news/2026-07-17/amazon-quick-agentic-sales-teammate/ 24 AI en 2026-07-17T00:00:00Z AWS: Amazon Quick Is an Agentic AI Teammate for Sales Teams That Automates CRM and Meeting Prep https://24-ai.news/en/news/2026-07-17/anthropic-claude-code-v2-1-212/ 24 AI en 2026-07-17T00:00:00Z Anthropic: Claude Code v2.1.212 brings a redesigned /fork, new /resume picker, and security fixes https://24-ai.news/en/news/2026-07-17/arxiv-llm-gpu-kernels-atrex/ 24 AI en 2026-07-17T00:00:00Z arXiv:2607.14541: Atrex-Bench Reveals LLM-Generated GPU Kernels Reach Only 10% of Target Hardware Speed https://24-ai.news/en/news/2026-07-17/arxiv-mcpevol-bench-agent-tools/ 24 AI en 2026-07-17T00:00:00Z arXiv:2607.14642: MCPEvol-Bench Shows Both GPT-5.4 and Claude Lose Accuracy When MCP Tools Change https://24-ai.news/en/news/2026-07-17/arxiv-physical-danger-prism-safety/ 24 AI en 2026-07-17T00:00:00Z arXiv:2607.15218: When Words Are Safe But Actions Kill — PRISM Separates Physical Danger from Text https://24-ai.news/en/news/2026-07-17/arxiv-pretraining-data-poisoning/ 24 AI en 2026-07-17T00:00:00Z arXiv:2607.15267: Poisoning Language Model Pretraining Data Through Computational Propaganda on Forums https://24-ai.news/en/news/2026-07-17/arxiv-transcoders-deception/ 24 AI en 2026-07-17T00:00:00Z arXiv:2607.14791: Transcoders Reveal How Deception Forms Within the Internal Circuits of the Qwen3-4B Language Model https://24-ai.news/en/news/2026-07-17/github-copilot-review-mobile-updates/ 24 AI en 2026-07-17T00:00:00Z GitHub: Copilot Code Review Gets a Firewall and Custom Runners, Mobile Gets a Cloud Agent Comment Fix https://24-ai.news/en/news/2026-07-17/github-cost-of-saying-yes/ 24 AI en 2026-07-17T00:00:00Z GitHub: The Cost of Writing Code Has Dropped, the Cost of Ownership Hasn't — A New Editorial on Engineering Economics https://24-ai.news/en/news/2026-07-17/nvidia-vera-rubin-post-training/ 24 AI en 2026-07-17T00:00:00Z NVIDIA: Vera Rubin Platform Trains the Largest AI Models with a Quarter of Blackwell-Generation GPUs Thanks to Codesign https://24-ai.news/en/news/2026-07-17/openai-ai-scorecard-roi-metrics/ 24 AI en 2026-07-17T00:00:00Z OpenAI: CFO Sarah Friar Proposes an AI Scorecard with Four Metrics to Measure ROI on Corporate AI Investment https://24-ai.news/en/news/2026-07-17/smartsheet-remote-mcp-server-aws/ 24 AI en 2026-07-17T00:00:00Z AWS: Smartsheet Saved 3 Billion Tokens by Building a Remote MCP Server with Bedrock and a Neptune Knowledge Graph https://24-ai.news/en/news/2026-07-17/vllm-ci-benchmarking-quality/ 24 AI en 2026-07-17T00:00:00Z vLLM: How 266 CI Jobs and Nightly Benchmarking Keep Production Quality for the Open-Source LLM Serving Engine https://24-ai.news/en/news/2026-07-17/xai-grok-4-5-api-release/ 24 AI en 2026-07-17T00:00:00Z xAI: Grok 4.5 Available on the API with Adjustable Reasoning-Effort Levels and Pricing from $2 per Token https://24-ai.news/de/news/2026-07-17/amazon-quick-agentic-sales-teammate/ 24 AI de 2026-07-17T00:00:00Z AWS: Amazon Quick ist ein agentischer KI-Kollege für Vertriebsteams, der CRM und Meeting-Vorbereitung automatisiert https://24-ai.news/de/news/2026-07-17/anthropic-claude-code-v2-1-212/ 24 AI de 2026-07-17T00:00:00Z Anthropic: Claude Code v2.1.212 bringt neu gestaltetes /fork, neuen /resume-Picker und Sicherheitskorrekturen https://24-ai.news/de/news/2026-07-17/arxiv-llm-gpu-kernels-atrex/ 24 AI de 2026-07-17T00:00:00Z arXiv:2607.14541: Atrex-Bench zeigt, dass LLM-generierte GPU-Kernel nur 10 % der angestrebten Hardware-Geschwindigkeit erreichen https://24-ai.news/de/news/2026-07-17/arxiv-mcpevol-bench-agent-tools/ 24 AI de 2026-07-17T00:00:00Z arXiv:2607.14642: MCPEvol-Bench zeigt, dass sowohl GPT-5.4 als auch Claude an Präzision verlieren, wenn sich MCP-Tools ändern https://24-ai.news/de/news/2026-07-17/arxiv-physical-danger-prism-safety/ 24 AI de 2026-07-17T00:00:00Z arXiv:2607.15218: Wenn Worte sicher sind, aber Handlungen töten — PRISM trennt physische Gefahr vom Text https://24-ai.news/de/news/2026-07-17/arxiv-pretraining-data-poisoning/ 24 AI de 2026-07-17T00:00:00Z arXiv:2607.15267: Vergiftung von Pretraining-Daten für Sprachmodelle durch computergestützte Propaganda in Foren https://24-ai.news/de/news/2026-07-17/arxiv-transcoders-deception/ 24 AI de 2026-07-17T00:00:00Z arXiv:2607.14791: Transcoder zeigen, wie Täuschung in den internen Schaltkreisen des Sprachmodells Qwen3-4B entsteht https://24-ai.news/de/news/2026-07-17/github-copilot-review-mobile-updates/ 24 AI de 2026-07-17T00:00:00Z GitHub: Copilot Code Review erhält Firewall und Custom Runner, Mobile erhält Cloud-Agent-Kommentarfix https://24-ai.news/de/news/2026-07-17/github-cost-of-saying-yes/ 24 AI de 2026-07-17T00:00:00Z GitHub: Die Kosten des Codeschreibens sind gesunken, die Kosten des Besitzes nicht — ein neues Editorial zur Ingenieurökonomie https://24-ai.news/de/news/2026-07-17/nvidia-vera-rubin-post-training/ 24 AI de 2026-07-17T00:00:00Z NVIDIA: Vera-Rubin-Plattform trainiert die größten KI-Modelle mit einem Viertel der Blackwell-GPUs dank Codesign https://24-ai.news/de/news/2026-07-17/openai-ai-scorecard-roi-metrics/ 24 AI de 2026-07-17T00:00:00Z OpenAI: CFO Sarah Friar schlägt AI Scorecard mit vier Kennzahlen zur Messung des ROI von KI-Investitionen vor https://24-ai.news/de/news/2026-07-17/smartsheet-remote-mcp-server-aws/ 24 AI de 2026-07-17T00:00:00Z AWS: Smartsheet spart 3 Milliarden Token durch den Bau eines Remote-MCP-Servers mit Bedrock und einem Neptune-Knowledge-Graph https://24-ai.news/de/news/2026-07-17/vllm-ci-benchmarking-quality/ 24 AI de 2026-07-17T00:00:00Z vLLM: Wie 266 CI-Jobs und nächtliches Benchmarking die Produktionsqualität der Open-Source-LLM-Serving-Engine bewahren https://24-ai.news/de/news/2026-07-17/xai-grok-4-5-api-release/ 24 AI de 2026-07-17T00:00:00Z xAI: Grok 4.5 auf der API verfügbar mit einstellbaren Reasoning-Effort-Stufen und Preisen ab 2 Dollar pro Token https://24-ai.news/zh/news/2026-07-17/amazon-quick-agentic-sales-teammate/ 24 AI zh-cn 2026-07-17T00:00:00Z AWS：Amazon Quick是为销售团队打造的智能体式AI同事，可自动化CRM与会议准备工作 https://24-ai.news/zh/news/2026-07-17/anthropic-claude-code-v2-1-212/ 24 AI zh-cn 2026-07-17T00:00:00Z Anthropic：Claude Code v2.1.212带来重新设计的/fork、全新/resume选择器与安全修复 https://24-ai.news/zh/news/2026-07-17/arxiv-llm-gpu-kernels-atrex/ 24 AI zh-cn 2026-07-17T00:00:00Z arXiv:2607.14541: Atrex-Bench揭示LLM生成的GPU内核仅达到目标硬件速度的10% https://24-ai.news/zh/news/2026-07-17/arxiv-mcpevol-bench-agent-tools/ 24 AI zh-cn 2026-07-17T00:00:00Z arXiv:2607.14642: MCPEvol-Bench显示GPT-5.4与Claude在MCP工具变化时都会失去精度 https://24-ai.news/zh/news/2026-07-17/arxiv-physical-danger-prism-safety/ 24 AI zh-cn 2026-07-17T00:00:00Z arXiv:2607.15218: 当言语安全但行动致命时——PRISM将物理危险从文本中区分出来 https://24-ai.news/zh/news/2026-07-17/arxiv-pretraining-data-poisoning/ 24 AI zh-cn 2026-07-17T00:00:00Z arXiv:2607.15267: 通过论坛上的计算宣传对语言模型预训练数据下毒 https://24-ai.news/zh/news/2026-07-17/arxiv-transcoders-deception/ 24 AI zh-cn 2026-07-17T00:00:00Z arXiv:2607.14791: 转码器揭示语言模型Qwen3-4B内部电路中欺骗是如何产生的 https://24-ai.news/zh/news/2026-07-17/github-copilot-review-mobile-updates/ 24 AI zh-cn 2026-07-17T00:00:00Z GitHub：Copilot代码评审获得防火墙与自定义运行器，手机端获得云代理评论修复功能 https://24-ai.news/zh/news/2026-07-17/github-cost-of-saying-yes/ 24 AI zh-cn 2026-07-17T00:00:00Z GitHub：编写代码的成本下降了，拥有代码的成本没有——一篇关于工程经济学的新社论 https://24-ai.news/zh/news/2026-07-17/nvidia-vera-rubin-post-training/ 24 AI zh-cn 2026-07-17T00:00:00Z NVIDIA：得益于协同设计，Vera Rubin平台以四分之一的Blackwell世代GPU数量训练最大规模的AI模型 https://24-ai.news/zh/news/2026-07-17/openai-ai-scorecard-roi-metrics/ 24 AI zh-cn 2026-07-17T00:00:00Z OpenAI：首席财务官Sarah Friar提出以四项指标衡量企业AI投资ROI的AI记分卡 https://24-ai.news/zh/news/2026-07-17/smartsheet-remote-mcp-server-aws/ 24 AI zh-cn 2026-07-17T00:00:00Z AWS：Smartsheet通过基于Bedrock与Neptune知识图谱构建远程MCP服务器节省30亿token https://24-ai.news/zh/news/2026-07-17/vllm-ci-benchmarking-quality/ 24 AI zh-cn 2026-07-17T00:00:00Z vLLM：266个CI任务与夜间基准测试如何守护这款开源LLM推理引擎的生产质量 https://24-ai.news/zh/news/2026-07-17/xai-grok-4-5-api-release/ 24 AI zh-cn 2026-07-17T00:00:00Z xAI：Grok 4.5登陆API，支持可调节的推理强度等级，价格为每百万输入token 2美元 https://24-ai.news/ja/news/2026-07-17/amazon-quick-agentic-sales-teammate/ 24 AI ja 2026-07-17T00:00:00Z AWS：Amazon Quickは営業チーム向けのエージェント型AIチームメイトで、CRMと会議準備を自動化 https://24-ai.news/ja/news/2026-07-17/anthropic-claude-code-v2-1-212/ 24 AI ja 2026-07-17T00:00:00Z Anthropic：Claude Code v2.1.212で/forkが刷新、新しい/resumeピッカーとセキュリティ修正を追加 https://24-ai.news/ja/news/2026-07-17/arxiv-llm-gpu-kernels-atrex/ 24 AI ja 2026-07-17T00:00:00Z arXiv:2607.14541：Atrex-Bench、LLM生成のGPUカーネルが目標ハードウェア速度のわずか10%にしか達しないことを明らかにする https://24-ai.news/ja/news/2026-07-17/arxiv-mcpevol-bench-agent-tools/ 24 AI ja 2026-07-17T00:00:00Z arXiv:2607.14642：MCPEvol-Bench、GPT-5.4とClaudeの両方がMCPツール変更時に精度を失うことを示す https://24-ai.news/ja/news/2026-07-17/arxiv-physical-danger-prism-safety/ 24 AI ja 2026-07-17T00:00:00Z arXiv:2607.15218：言葉は安全でも行動が命を奪うとき——PRISMは物理的危険をテキストから切り分ける https://24-ai.news/ja/news/2026-07-17/arxiv-pretraining-data-poisoning/ 24 AI ja 2026-07-17T00:00:00Z arXiv:2607.15267：フォーラムにおけるコンピュテーショナル・プロパガンダを通じた言語モデル事前学習データの毒化 https://24-ai.news/ja/news/2026-07-17/arxiv-transcoders-deception/ 24 AI ja 2026-07-17T00:00:00Z arXiv:2607.14791：トランスコーダーが言語モデルQwen3-4Bの内部回路における欺瞞の発生メカニズムを明らかにする https://24-ai.news/ja/news/2026-07-17/github-copilot-review-mobile-updates/ 24 AI ja 2026-07-17T00:00:00Z GitHub：Copilotコードレビューにファイアウォールとカスタムランナーを追加、モバイルにはクラウドエージェントによるコメント修正機能 https://24-ai.news/ja/news/2026-07-17/github-cost-of-saying-yes/ 24 AI ja 2026-07-17T00:00:00Z GitHub：コードを書くコストは下がったが、所有するコストは下がっていない——エンジニアリング経済学に関する新しい論説 https://24-ai.news/ja/news/2026-07-17/nvidia-vera-rubin-post-training/ 24 AI ja 2026-07-17T00:00:00Z NVIDIA：コデザインによりVera RubinプラットフォームがBlackwell世代の4分の1のGPUで最大級のAIモデルを訓練 https://24-ai.news/ja/news/2026-07-17/openai-ai-scorecard-roi-metrics/ 24 AI ja 2026-07-17T00:00:00Z OpenAI：CFOサラ・フライアーが企業のAI投資ROIを測る4指標のAIスコアカードを提案 https://24-ai.news/ja/news/2026-07-17/smartsheet-remote-mcp-server-aws/ 24 AI ja 2026-07-17T00:00:00Z AWS：Smartsheetが、BedrockとNeptuneナレッジグラフによるリモートMCPサーバー構築で30億トークンを節約 https://24-ai.news/ja/news/2026-07-17/vllm-ci-benchmarking-quality/ 24 AI ja 2026-07-17T00:00:00Z vLLM：266のCIジョブと夜間ベンチマークがオープンソースLLM推論エンジンの本番品質をどう守るか https://24-ai.news/ja/news/2026-07-17/xai-grok-4-5-api-release/ 24 AI ja 2026-07-17T00:00:00Z xAI：Grok 4.5がAPIで利用可能に、調整可能な推論強度と入力100万トークンあたり2ドルの価格 https://24-ai.news/ko/news/2026-07-17/amazon-quick-agentic-sales-teammate/ 24 AI ko 2026-07-17T00:00:00Z AWS: Amazon Quick, CRM과 회의 준비를 자동화하는 영업팀용 에이전틱 AI 동료 https://24-ai.news/ko/news/2026-07-17/anthropic-claude-code-v2-1-212/ 24 AI ko 2026-07-17T00:00:00Z Anthropic: Claude Code v2.1.212, 재설계된 /fork와 새로운 /resume 피커, 보안 수정 제공 https://24-ai.news/ko/news/2026-07-17/arxiv-llm-gpu-kernels-atrex/ 24 AI ko 2026-07-17T00:00:00Z arXiv:2607.14541: Atrex-Bench, LLM이 생성한 GPU 커널이 목표 하드웨어 속도의 단 10%에 그친다는 것을 밝히다 https://24-ai.news/ko/news/2026-07-17/arxiv-mcpevol-bench-agent-tools/ 24 AI ko 2026-07-17T00:00:00Z arXiv:2607.14642: MCPEvol-Bench, GPT-5.4와 Claude 모두 MCP 도구 변경 시 정확도를 잃는다는 것을 보여주다 https://24-ai.news/ko/news/2026-07-17/arxiv-physical-danger-prism-safety/ 24 AI ko 2026-07-17T00:00:00Z arXiv:2607.15218: 말은 안전하지만 행동이 사람을 죽일 때 — PRISM은 물리적 위험을 텍스트에서 분리한다 https://24-ai.news/ko/news/2026-07-17/arxiv-pretraining-data-poisoning/ 24 AI ko 2026-07-17T00:00:00Z arXiv:2607.15267: 포럼을 통한 계산 선전으로 언어 모델 사전학습 데이터 오염시키기 https://24-ai.news/ko/news/2026-07-17/arxiv-transcoders-deception/ 24 AI ko 2026-07-17T00:00:00Z arXiv:2607.14791: 트랜스코더가 언어 모델 Qwen3-4B의 내부 회로에서 거짓말이 생성되는 과정을 밝히다 https://24-ai.news/ko/news/2026-07-17/github-copilot-review-mobile-updates/ 24 AI ko 2026-07-17T00:00:00Z GitHub: Copilot 코드 리뷰에 방화벽과 커스텀 러너 추가, 모바일에는 클라우드 에이전트 댓글 수정 기능 도입 https://24-ai.news/ko/news/2026-07-17/github-cost-of-saying-yes/ 24 AI ko 2026-07-17T00:00:00Z GitHub: 코드 작성 비용은 떨어졌지만 소유 비용은 그대로다 — 엔지니어링 경제학에 관한 새 사설 https://24-ai.news/ko/news/2026-07-17/nvidia-vera-rubin-post-training/ 24 AI ko 2026-07-17T00:00:00Z NVIDIA: 코디자인 덕분에 Vera Rubin 플랫폼이 Blackwell 세대 GPU의 4분의 1만으로 최대 규모 AI 모델 훈련 https://24-ai.news/ko/news/2026-07-17/openai-ai-scorecard-roi-metrics/ 24 AI ko 2026-07-17T00:00:00Z OpenAI: CFO 사라 프리어, 기업 AI 투자 ROI 측정을 위한 4가지 지표의 AI 스코어카드 제안 https://24-ai.news/ko/news/2026-07-17/smartsheet-remote-mcp-server-aws/ 24 AI ko 2026-07-17T00:00:00Z AWS: Smartsheet, Bedrock와 Neptune 지식 그래프 기반 원격 MCP 서버 구축으로 30억 토큰 절약 https://24-ai.news/ko/news/2026-07-17/vllm-ci-benchmarking-quality/ 24 AI ko 2026-07-17T00:00:00Z vLLM: 266개의 CI 작업과 야간 벤치마킹이 오픈소스 LLM 추론 엔진의 프로덕션 품질을 지키는 방법 https://24-ai.news/ko/news/2026-07-17/xai-grok-4-5-api-release/ 24 AI ko 2026-07-17T00:00:00Z xAI: Grok 4.5, 조정 가능한 추론 강도 등급과 토큰당 2달러 가격으로 API에 출시 https://24-ai.news/hr/news/2026-07-16/amd-aims-2-2-multi-accelerator-inference/ 24 AI hr 2026-07-16T00:00:00Z AMD: AIMs 2.2 donosi inference-mikroservise za Instinct, EPYC i Radeon https://24-ai.news/hr/news/2026-07-16/amd-profiling-cursor-claude-kernel-opt/ 24 AI hr 2026-07-16T00:00:00Z AMD: AI kod-asistent (Cursor + Claude Opus 4.7) ubrzao GPU kernel 28,3× na MI250 https://24-ai.news/hr/news/2026-07-16/arxiv-ai-advice-suppresses-i-dont-know/ 24 AI hr 2026-07-16T00:00:00Z arXiv:2607.13562: AI savjeti gase spremnost ljudi da kažu 'ne znam' — čak i kad su pogrešni https://24-ai.news/hr/news/2026-07-16/arxiv-deep-interaction-reasoning-editing/ 24 AI hr 2026-07-16T00:00:00Z arXiv:2607.14049: 'Deep Interaction' — ispravljanje reasoning-lanca podiže korekciju +25%, štedi 40% tokena https://24-ai.news/hr/news/2026-07-16/arxiv-self-improving-agents-survey/ 24 AI hr 2026-07-16T00:00:00Z arXiv:2607.13104: opsežan pregled samopoboljšavajućih AI agenata (uz Jürgena Schmidhubera) https://24-ai.news/hr/news/2026-07-16/arxiv-theory-level-autoformalization/ 24 AI hr 2026-07-16T00:00:00Z arXiv:2607.13292: autoformalizacija na razini teorije — cijele matematičke baze umjesto izoliranih iskaza (ICML 2026) https://24-ai.news/hr/news/2026-07-16/aws-grok-4-3-amazon-bedrock/ 24 AI hr 2026-07-16T00:00:00Z AWS: xAI-jev Grok 4.3 sada dostupan na Amazon Bedrocku https://24-ai.news/hr/news/2026-07-16/deepmind-bioresilience-biosecurity-ai/ 24 AI hr 2026-07-16T00:00:00Z Google DeepMind: pristup 'biorezilijenciji' — AI protiv bioloških prijetnji https://24-ai.news/hr/news/2026-07-16/google-search-connected-apps-vids-omni/ 24 AI hr 2026-07-16T00:00:00Z Google: AI Mode povezuje aplikacije u Searchu + Gemini Omni i avatari u Vidsu https://24-ai.news/hr/news/2026-07-16/langchain-openwiki-okf-agent-docs/ 24 AI hr 2026-07-16T00:00:00Z LangChain: OpenWiki 0.2 uvodi OKF — Google Cloudov standard za agentsku dokumentaciju koda https://24-ai.news/hr/news/2026-07-16/oecd-haip-ai-transparency-framework/ 24 AI hr 2026-07-16T00:00:00Z OECD: HAIP okvir pretvara transparentnost AI-ja iz tereta u konkurentsku prednost https://24-ai.news/hr/news/2026-07-16/openai-teens-safe-ai-access/ 24 AI hr 2026-07-16T00:00:00Z OpenAI: zašto tinejdžeri zaslužuju pristup sigurnom AI-ju https://24-ai.news/hr/news/2026-07-16/sakana-nvidia-nemotron-fugu-agents/ 24 AI hr 2026-07-16T00:00:00Z Sakana AI i NVIDIA: Nemotron open-modeli u multi-agentnom sustavu Fugu https://24-ai.news/en/news/2026-07-16/amd-aims-2-2-multi-accelerator-inference/ 24 AI en 2026-07-16T00:00:00Z AMD: AIMs 2.2 Brings Inference Microservices for Instinct, EPYC, and Radeon https://24-ai.news/en/news/2026-07-16/amd-profiling-cursor-claude-kernel-opt/ 24 AI en 2026-07-16T00:00:00Z AMD: AI Code Assistant (Cursor + Claude Opus 4.7) Speeds Up GPU Kernel 28.3× on MI250 https://24-ai.news/en/news/2026-07-16/arxiv-ai-advice-suppresses-i-dont-know/ 24 AI en 2026-07-16T00:00:00Z arXiv:2607.13562: AI Advice Suppresses People's Willingness to Say 'I Don't Know' — Even When Wrong https://24-ai.news/en/news/2026-07-16/arxiv-deep-interaction-reasoning-editing/ 24 AI en 2026-07-16T00:00:00Z arXiv:2607.14049: 'Deep Interaction' — Editing the Reasoning Chain Boosts Correction by +25%, Saves 40% of Tokens https://24-ai.news/en/news/2026-07-16/arxiv-self-improving-agents-survey/ 24 AI en 2026-07-16T00:00:00Z arXiv:2607.13104: Comprehensive Survey of Self-Improving AI Agents (With Jürgen Schmidhuber) https://24-ai.news/en/news/2026-07-16/arxiv-theory-level-autoformalization/ 24 AI en 2026-07-16T00:00:00Z arXiv:2607.13292: Theory-Level Autoformalization — Entire Mathematical Bases Instead of Isolated Statements (ICML 2026) https://24-ai.news/en/news/2026-07-16/aws-grok-4-3-amazon-bedrock/ 24 AI en 2026-07-16T00:00:00Z AWS: xAI's Grok 4.3 Now Available on Amazon Bedrock https://24-ai.news/en/news/2026-07-16/deepmind-bioresilience-biosecurity-ai/ 24 AI en 2026-07-16T00:00:00Z Google DeepMind: A 'Bioresilience' Approach — AI Against Biological Threats https://24-ai.news/en/news/2026-07-16/google-search-connected-apps-vids-omni/ 24 AI en 2026-07-16T00:00:00Z Google: AI Mode Connects Apps in Search + Gemini Omni and Avatars in Vids https://24-ai.news/en/news/2026-07-16/langchain-openwiki-okf-agent-docs/ 24 AI en 2026-07-16T00:00:00Z LangChain: OpenWiki 0.2 Introduces OKF — Google Cloud's Standard for Agent Code Documentation https://24-ai.news/en/news/2026-07-16/oecd-haip-ai-transparency-framework/ 24 AI en 2026-07-16T00:00:00Z OECD: HAIP Framework Turns AI Transparency From Burden Into Competitive Advantage https://24-ai.news/en/news/2026-07-16/openai-teens-safe-ai-access/ 24 AI en 2026-07-16T00:00:00Z OpenAI: Why Teens Deserve Access to Safe AI https://24-ai.news/en/news/2026-07-16/sakana-nvidia-nemotron-fugu-agents/ 24 AI en 2026-07-16T00:00:00Z Sakana AI and NVIDIA: Nemotron Open Models in the Fugu Multi-Agent System https://24-ai.news/de/news/2026-07-16/amd-aims-2-2-multi-accelerator-inference/ 24 AI de 2026-07-16T00:00:00Z AMD: AIMs 2.2 bringt Inference-Mikroservices für Instinct, EPYC und Radeon https://24-ai.news/de/news/2026-07-16/amd-profiling-cursor-claude-kernel-opt/ 24 AI de 2026-07-16T00:00:00Z AMD: KI-Code-Assistent (Cursor + Claude Opus 4.7) beschleunigt GPU-Kernel 28,3× auf MI250 https://24-ai.news/de/news/2026-07-16/arxiv-ai-advice-suppresses-i-dont-know/ 24 AI de 2026-07-16T00:00:00Z arXiv:2607.13562: KI-Ratschläge unterdrücken die Bereitschaft, "Ich weiß es nicht" zu sagen — selbst bei falschen Antworten https://24-ai.news/de/news/2026-07-16/arxiv-deep-interaction-reasoning-editing/ 24 AI de 2026-07-16T00:00:00Z arXiv:2607.14049: "Deep Interaction" — Reasoning-Chain bearbeiten steigert Korrekturen um +25 %, spart 40 % Token https://24-ai.news/de/news/2026-07-16/arxiv-self-improving-agents-survey/ 24 AI de 2026-07-16T00:00:00Z arXiv:2607.13104: Umfassender Survey zu selbstverbessernden KI-Agenten (mit Jürgen Schmidhuber) https://24-ai.news/de/news/2026-07-16/arxiv-theory-level-autoformalization/ 24 AI de 2026-07-16T00:00:00Z arXiv:2607.13292: Autoformalisierung auf Theorieebene — Ganze mathematische Basen statt isolierter Aussagen (ICML 2026) https://24-ai.news/de/news/2026-07-16/aws-grok-4-3-amazon-bedrock/ 24 AI de 2026-07-16T00:00:00Z AWS: xAIs Grok 4.3 jetzt auf Amazon Bedrock verfügbar https://24-ai.news/de/news/2026-07-16/deepmind-bioresilience-biosecurity-ai/ 24 AI de 2026-07-16T00:00:00Z Google DeepMind: Ansatz zur "Bioresilienz" — KI gegen biologische Bedrohungen https://24-ai.news/de/news/2026-07-16/google-search-connected-apps-vids-omni/ 24 AI de 2026-07-16T00:00:00Z Google: AI Mode verbindet Apps in der Suche + Gemini Omni und Avatare in Vids https://24-ai.news/de/news/2026-07-16/langchain-openwiki-okf-agent-docs/ 24 AI de 2026-07-16T00:00:00Z LangChain: OpenWiki 0.2 führt OKF ein — Google Clouds Standard für agentische Code-Dokumentation https://24-ai.news/de/news/2026-07-16/oecd-haip-ai-transparency-framework/ 24 AI de 2026-07-16T00:00:00Z OECD: HAIP-Rahmen macht KI-Transparenz zum Wettbewerbsvorteil statt zur Compliance-Last https://24-ai.news/de/news/2026-07-16/openai-teens-safe-ai-access/ 24 AI de 2026-07-16T00:00:00Z OpenAI: Warum Teenager Zugang zu sicherer KI verdienen https://24-ai.news/de/news/2026-07-16/sakana-nvidia-nemotron-fugu-agents/ 24 AI de 2026-07-16T00:00:00Z Sakana AI und NVIDIA: Nemotron Open-Modelle im Multi-Agenten-System Fugu https://24-ai.news/zh/news/2026-07-16/amd-aims-2-2-multi-accelerator-inference/ 24 AI zh-cn 2026-07-16T00:00:00Z AMD：AIMs 2.2带来面向Instinct、EPYC和Radeon的推理微服务 https://24-ai.news/zh/news/2026-07-16/amd-profiling-cursor-claude-kernel-opt/ 24 AI zh-cn 2026-07-16T00:00:00Z AMD：AI代码助手（Cursor + Claude Opus 4.7）将MI250 GPU内核加速28.3倍 https://24-ai.news/zh/news/2026-07-16/arxiv-ai-advice-suppresses-i-dont-know/ 24 AI zh-cn 2026-07-16T00:00:00Z arXiv:2607.13562：AI建议抑制人们说「我不知道」的意愿——即便AI答案是错的 https://24-ai.news/zh/news/2026-07-16/arxiv-deep-interaction-reasoning-editing/ 24 AI zh-cn 2026-07-16T00:00:00Z arXiv:2607.14049：「深度交互」——编辑推理链使纠正成功率提升+25%，节省40%令牌 https://24-ai.news/zh/news/2026-07-16/arxiv-self-improving-agents-survey/ 24 AI zh-cn 2026-07-16T00:00:00Z arXiv:2607.13104：自我改进AI智能体综述（含Jürgen Schmidhuber） https://24-ai.news/zh/news/2026-07-16/arxiv-theory-level-autoformalization/ 24 AI zh-cn 2026-07-16T00:00:00Z arXiv:2607.13292：理论层面的自动形式化——整个数学理论体系而非孤立命题（ICML 2026） https://24-ai.news/zh/news/2026-07-16/aws-grok-4-3-amazon-bedrock/ 24 AI zh-cn 2026-07-16T00:00:00Z AWS：xAI的Grok 4.3现已在Amazon Bedrock上可用 https://24-ai.news/zh/news/2026-07-16/deepmind-bioresilience-biosecurity-ai/ 24 AI zh-cn 2026-07-16T00:00:00Z Google DeepMind：「生物韧性」方法——AI应对生物威胁 https://24-ai.news/zh/news/2026-07-16/google-search-connected-apps-vids-omni/ 24 AI zh-cn 2026-07-16T00:00:00Z Google：AI Mode在搜索中连接应用程序 + Gemini Omni与Vids中的虚拟形象 https://24-ai.news/zh/news/2026-07-16/langchain-openwiki-okf-agent-docs/ 24 AI zh-cn 2026-07-16T00:00:00Z LangChain：OpenWiki 0.2引入OKF——Google Cloud的智能体代码文档标准 https://24-ai.news/zh/news/2026-07-16/oecd-haip-ai-transparency-framework/ 24 AI zh-cn 2026-07-16T00:00:00Z OECD：HAIP框架将AI透明度从合规负担转化为竞争优势 https://24-ai.news/zh/news/2026-07-16/openai-teens-safe-ai-access/ 24 AI zh-cn 2026-07-16T00:00:00Z OpenAI：青少年为何值得获得安全AI的访问权 https://24-ai.news/zh/news/2026-07-16/sakana-nvidia-nemotron-fugu-agents/ 24 AI zh-cn 2026-07-16T00:00:00Z Sakana AI与NVIDIA：Nemotron开源模型集成至多智能体系统Fugu https://24-ai.news/ja/news/2026-07-16/amd-aims-2-2-multi-accelerator-inference/ 24 AI ja 2026-07-16T00:00:00Z AMD：AIMs 2.2がInstinct・EPYC・Radeon向け推論マイクロサービスを提供 https://24-ai.news/ja/news/2026-07-16/amd-profiling-cursor-claude-kernel-opt/ 24 AI ja 2026-07-16T00:00:00Z AMD：AIコードアシスタント（Cursor + Claude Opus 4.7）がMI250でGPUカーネルを28.3倍高速化 https://24-ai.news/ja/news/2026-07-16/arxiv-ai-advice-suppresses-i-dont-know/ 24 AI ja 2026-07-16T00:00:00Z arXiv:2607.13562：AIのアドバイスは「わからない」と言う意欲を消す——回答が誤りでも https://24-ai.news/ja/news/2026-07-16/arxiv-deep-interaction-reasoning-editing/ 24 AI ja 2026-07-16T00:00:00Z arXiv:2607.14049：「Deep Interaction」——推論チェーン編集が修正成功率+25%、トークン40%節約 https://24-ai.news/ja/news/2026-07-16/arxiv-self-improving-agents-survey/ 24 AI ja 2026-07-16T00:00:00Z arXiv:2607.13104：自己改善型AIエージェントの包括的サーベイ（Jürgen Schmidhuber参加） https://24-ai.news/ja/news/2026-07-16/arxiv-theory-level-autoformalization/ 24 AI ja 2026-07-16T00:00:00Z arXiv:2607.13292：理論レベルの自動形式化——孤立した命題ではなく数学理論全体を対象に（ICML 2026） https://24-ai.news/ja/news/2026-07-16/aws-grok-4-3-amazon-bedrock/ 24 AI ja 2026-07-16T00:00:00Z AWS：xAIのGrok 4.3がAmazon Bedrockで利用可能に https://24-ai.news/ja/news/2026-07-16/deepmind-bioresilience-biosecurity-ai/ 24 AI ja 2026-07-16T00:00:00Z Google DeepMind：「バイオレジリエンス」アプローチ——AIが生物学的脅威に対抗 https://24-ai.news/ja/news/2026-07-16/google-search-connected-apps-vids-omni/ 24 AI ja 2026-07-16T00:00:00Z Google：AI ModeがSearchでアプリを連携 + VidsにGemini Omniとアバター https://24-ai.news/ja/news/2026-07-16/langchain-openwiki-okf-agent-docs/ 24 AI ja 2026-07-16T00:00:00Z LangChain：OpenWiki 0.2がOKFを導入——Google Cloudのエージェント向けコードドキュメント標準 https://24-ai.news/ja/news/2026-07-16/oecd-haip-ai-transparency-framework/ 24 AI ja 2026-07-16T00:00:00Z OECD：HAIPフレームワークがAI透明性をコンプライアンス負担から競争優位へ転換 https://24-ai.news/ja/news/2026-07-16/openai-teens-safe-ai-access/ 24 AI ja 2026-07-16T00:00:00Z OpenAI：ティーンエイジャーが安全なAIへのアクセスを受けるべき理由 https://24-ai.news/ja/news/2026-07-16/sakana-nvidia-nemotron-fugu-agents/ 24 AI ja 2026-07-16T00:00:00Z Sakana AIとNVIDIA：NemotronオープンモデルをマルチエージェントシステムのFuguに統合 https://24-ai.news/ko/news/2026-07-16/amd-aims-2-2-multi-accelerator-inference/ 24 AI ko 2026-07-16T00:00:00Z AMD: AIMs 2.2, Instinct·EPYC·Radeon 위한 추론 마이크로서비스 제공 https://24-ai.news/ko/news/2026-07-16/amd-profiling-cursor-claude-kernel-opt/ 24 AI ko 2026-07-16T00:00:00Z AMD: AI 코드 어시스턴트(Cursor + Claude Opus 4.7)가 MI250에서 GPU 커널 28.3배 가속 https://24-ai.news/ko/news/2026-07-16/arxiv-ai-advice-suppresses-i-dont-know/ 24 AI ko 2026-07-16T00:00:00Z arXiv:2607.13562: AI 조언은 『모르겠다』는 의향을 억누른다—답이 틀려도 https://24-ai.news/ko/news/2026-07-16/arxiv-deep-interaction-reasoning-editing/ 24 AI ko 2026-07-16T00:00:00Z arXiv:2607.14049: 『Deep Interaction』—추론 체인 수정이 교정 성공률 +25%, 토큰 40% 절약 https://24-ai.news/ko/news/2026-07-16/arxiv-self-improving-agents-survey/ 24 AI ko 2026-07-16T00:00:00Z arXiv:2607.13104: 자기 개선 AI 에이전트 종합 설문 (Jürgen Schmidhuber 참여) https://24-ai.news/ko/news/2026-07-16/arxiv-theory-level-autoformalization/ 24 AI ko 2026-07-16T00:00:00Z arXiv:2607.13292: 이론 수준의 자동 형식화—고립된 명제가 아닌 전체 수학 이론 (ICML 2026) https://24-ai.news/ko/news/2026-07-16/aws-grok-4-3-amazon-bedrock/ 24 AI ko 2026-07-16T00:00:00Z AWS: xAI의 Grok 4.3, Amazon Bedrock에서 이용 가능 https://24-ai.news/ko/news/2026-07-16/deepmind-bioresilience-biosecurity-ai/ 24 AI ko 2026-07-16T00:00:00Z Google DeepMind: 『바이오레질리언스』 접근법—AI가 생물학적 위협에 대응하다 https://24-ai.news/ko/news/2026-07-16/google-search-connected-apps-vids-omni/ 24 AI ko 2026-07-16T00:00:00Z Google: AI Mode가 검색에서 앱 연결 + Vids에 Gemini Omni와 아바타 https://24-ai.news/ko/news/2026-07-16/langchain-openwiki-okf-agent-docs/ 24 AI ko 2026-07-16T00:00:00Z LangChain: OpenWiki 0.2가 OKF 도입—Google Cloud의 에이전트 코드 문서 표준 https://24-ai.news/ko/news/2026-07-16/oecd-haip-ai-transparency-framework/ 24 AI ko 2026-07-16T00:00:00Z OECD: HAIP 프레임워크가 AI 투명성을 컴플라이언스 부담에서 경쟁 우위로 전환 https://24-ai.news/ko/news/2026-07-16/openai-teens-safe-ai-access/ 24 AI ko 2026-07-16T00:00:00Z OpenAI: 청소년이 안전한 AI에 접근할 자격이 있는 이유 https://24-ai.news/ko/news/2026-07-16/sakana-nvidia-nemotron-fugu-agents/ 24 AI ko 2026-07-16T00:00:00Z Sakana AI와 NVIDIA: Nemotron 오픈 모델을 멀티 에이전트 시스템 Fugu에 통합 https://24-ai.news/hr/news/2026-07-15/amd-hipvs-agentic-rag-vector-search/ 24 AI hr 2026-07-15T00:00:00Z AMD: hipVS — GPU vektorska pretraga za Instinct, cuVS-kompatibilna, s demoom agentnog RAG-a https://24-ai.news/hr/news/2026-07-15/arxiv-cbrn-uplift-frontier-llm-eval/ 24 AI hr 2026-07-15T00:00:00Z arXiv:2607.12200: okvir za mjerenje CBRN 'uplifta' kod frontier modela — potvrđen rizik samo u radiološkoj domeni https://24-ai.news/hr/news/2026-07-15/arxiv-e3-agents-task-complexity-estimate/ 24 AI hr 2026-07-15T00:00:00Z arXiv:2607.13034: E3 okvir — agenti procjenjuju složenost zadatka i troše 91% manje tokena https://24-ai.news/hr/news/2026-07-15/arxiv-fim-coding-agent-mid-training/ 24 AI hr 2026-07-15T00:00:00Z arXiv:2607.12463: funkcijski svjestan FIM mid-trening podiže coding-agente do +5,4 na SWE-Benchu https://24-ai.news/hr/news/2026-07-15/arxiv-pm-bench-prospective-memory-agents/ 24 AI hr 2026-07-15T00:00:00Z arXiv:2607.12385: PM-Bench mjeri 'prospektivnu memoriju' agenata — najbolji GPT-5.4 tek 65,1% F1 https://24-ai.news/hr/news/2026-07-15/cncf-hami-gpu-virtualization-incubating/ 24 AI hr 2026-07-15T00:00:00Z CNCF: HAMi postao Incubating projekt — virtualizacija GPU-a za Kubernetes AI opterećenja https://24-ai.news/hr/news/2026-07-15/google-diffusion-model-creativity/ 24 AI hr 2026-07-15T00:00:00Z Google Research: kreativnost difuzijskih modela objašnjena kao 'zaglađivanje' score funkcije https://24-ai.news/hr/news/2026-07-15/langchain-agents-sandboxed-computer/ 24 AI hr 2026-07-15T00:00:00Z LangChain: zašto AI agentima treba vlastito izolirano računalo (sandbox) https://24-ai.news/hr/news/2026-07-15/lfai-context-layer-docling-doclang/ 24 AI hr 2026-07-15T00:00:00Z LF AI & Data: 'kontekst je novo usko grlo' — otvorena infrastruktura Docling i DocLang https://24-ai.news/hr/news/2026-07-15/openai-state-federal-ai-safety-action/ 24 AI hr 2026-07-15T00:00:00Z OpenAI: zagovaranje 'obrnutog federalizma' za regulaciju AI-ja u SAD-u https://24-ai.news/hr/news/2026-07-15/pytorch-triton-plugin-extensions-tlx/ 24 AI hr 2026-07-15T00:00:00Z PyTorch: Triton 3.7 donosi plugin-proširenja i TLX — do +15% throughputa na AMD MI350 https://24-ai.news/hr/news/2026-07-15/vllm-tml-inkling-trillion-param-day0/ 24 AI hr 2026-07-15T00:00:00Z vLLM: Day-0 podrška za TML Inkling — bilijun-parametarski model do 380 tokena/s na GB200 https://24-ai.news/en/news/2026-07-15/amd-hipvs-agentic-rag-vector-search/ 24 AI en 2026-07-15T00:00:00Z AMD: hipVS — GPU vector search for Instinct, cuVS-compatible, with agentic RAG demo https://24-ai.news/en/news/2026-07-15/arxiv-cbrn-uplift-frontier-llm-eval/ 24 AI en 2026-07-15T00:00:00Z arXiv:2607.12200: framework for measuring CBRN 'uplift' in frontier models — material risk confirmed only in radiological domain https://24-ai.news/en/news/2026-07-15/arxiv-e3-agents-task-complexity-estimate/ 24 AI en 2026-07-15T00:00:00Z arXiv:2607.13034: E3 framework — agents estimate task complexity and use 91% fewer tokens https://24-ai.news/en/news/2026-07-15/arxiv-fim-coding-agent-mid-training/ 24 AI en 2026-07-15T00:00:00Z arXiv:2607.12463: function-aware FIM mid-training boosts coding agents up to +5.4 on SWE-Bench https://24-ai.news/en/news/2026-07-15/arxiv-pm-bench-prospective-memory-agents/ 24 AI en 2026-07-15T00:00:00Z arXiv:2607.12385: PM-Bench measures 'prospective memory' in agents — best GPT-5.4 scores only 65.1% F1 https://24-ai.news/en/news/2026-07-15/cncf-hami-gpu-virtualization-incubating/ 24 AI en 2026-07-15T00:00:00Z CNCF: HAMi becomes Incubating project — GPU virtualization for Kubernetes AI workloads https://24-ai.news/en/news/2026-07-15/google-diffusion-model-creativity/ 24 AI en 2026-07-15T00:00:00Z Google Research: creativity of diffusion models explained as 'score smoothing' https://24-ai.news/en/news/2026-07-15/langchain-agents-sandboxed-computer/ 24 AI en 2026-07-15T00:00:00Z LangChain: why AI agents need their own isolated computer (sandbox) https://24-ai.news/en/news/2026-07-15/lfai-context-layer-docling-doclang/ 24 AI en 2026-07-15T00:00:00Z LF AI & Data: 'context is the new bottleneck' — open infrastructure Docling and DocLang https://24-ai.news/en/news/2026-07-15/openai-state-federal-ai-safety-action/ 24 AI en 2026-07-15T00:00:00Z OpenAI: advocating 'reverse federalism' for AI regulation in the US https://24-ai.news/en/news/2026-07-15/pytorch-triton-plugin-extensions-tlx/ 24 AI en 2026-07-15T00:00:00Z PyTorch: Triton 3.7 brings plugin extensions and TLX — up to +15% throughput on AMD MI350 https://24-ai.news/en/news/2026-07-15/vllm-tml-inkling-trillion-param-day0/ 24 AI en 2026-07-15T00:00:00Z vLLM: Day-0 support for TML Inkling — trillion-parameter model up to 380 tokens/s on GB200 https://24-ai.news/de/news/2026-07-15/amd-hipvs-agentic-rag-vector-search/ 24 AI de 2026-07-15T00:00:00Z AMD: hipVS — GPU-Vektorsuche für Instinct, cuVS-kompatibel, mit agentischem RAG-Demo https://24-ai.news/de/news/2026-07-15/arxiv-cbrn-uplift-frontier-llm-eval/ 24 AI de 2026-07-15T00:00:00Z arXiv:2607.12200: Framework zur Messung von CBRN-„Uplift" bei Frontier-Modellen — materielles Risiko nur in der radiologischen Domäne bestätigt https://24-ai.news/de/news/2026-07-15/arxiv-e3-agents-task-complexity-estimate/ 24 AI de 2026-07-15T00:00:00Z arXiv:2607.13034: E3-Framework — Agenten schätzen Aufgabenkomplexität und verbrauchen 91 % weniger Token https://24-ai.news/de/news/2026-07-15/arxiv-fim-coding-agent-mid-training/ 24 AI de 2026-07-15T00:00:00Z arXiv:2607.12463: funktionsbewusstes FIM-Mid-Training verbessert Coding-Agenten um bis zu +5,4 auf SWE-Bench https://24-ai.news/de/news/2026-07-15/arxiv-pm-bench-prospective-memory-agents/ 24 AI de 2026-07-15T00:00:00Z arXiv:2607.12385: PM-Bench misst „prospektives Gedächtnis" von Agenten — bestes Modell GPT-5.4 erreicht nur 65,1 % F1 https://24-ai.news/de/news/2026-07-15/cncf-hami-gpu-virtualization-incubating/ 24 AI de 2026-07-15T00:00:00Z CNCF: HAMi wird Incubating-Projekt — GPU-Virtualisierung für Kubernetes-KI-Workloads https://24-ai.news/de/news/2026-07-15/google-diffusion-model-creativity/ 24 AI de 2026-07-15T00:00:00Z Google Research: Kreativität von Diffusionsmodellen als „Score-Glättung" erklärt https://24-ai.news/de/news/2026-07-15/langchain-agents-sandboxed-computer/ 24 AI de 2026-07-15T00:00:00Z LangChain: Warum KI-Agenten einen eigenen isolierten Computer (Sandbox) benötigen https://24-ai.news/de/news/2026-07-15/lfai-context-layer-docling-doclang/ 24 AI de 2026-07-15T00:00:00Z LF AI & Data: „Kontext ist der neue Engpass" — offene Infrastruktur Docling und DocLang https://24-ai.news/de/news/2026-07-15/openai-state-federal-ai-safety-action/ 24 AI de 2026-07-15T00:00:00Z OpenAI: Plädoyer für „umgekehrten Föderalismus" bei der KI-Regulierung in den USA https://24-ai.news/de/news/2026-07-15/pytorch-triton-plugin-extensions-tlx/ 24 AI de 2026-07-15T00:00:00Z PyTorch: Triton 3.7 bringt Plugin-Erweiterungen und TLX — bis zu +15 % Durchsatz auf AMD MI350 https://24-ai.news/de/news/2026-07-15/vllm-tml-inkling-trillion-param-day0/ 24 AI de 2026-07-15T00:00:00Z vLLM: Day-0-Unterstützung für TML Inkling — Billion-Parameter-Modell bis zu 380 Token/s auf GB200 https://24-ai.news/zh/news/2026-07-15/amd-hipvs-agentic-rag-vector-search/ 24 AI zh-cn 2026-07-15T00:00:00Z AMD：hipVS——面向Instinct的GPU向量检索库，兼容cuVS API，并附智能体RAG演示 https://24-ai.news/zh/news/2026-07-15/arxiv-cbrn-uplift-frontier-llm-eval/ 24 AI zh-cn 2026-07-15T00:00:00Z arXiv:2607.12200：评估前沿模型 CBRN「提升」风险的 TEC 框架——仅在放射性领域确认风险 https://24-ai.news/zh/news/2026-07-15/arxiv-e3-agents-task-complexity-estimate/ 24 AI zh-cn 2026-07-15T00:00:00Z arXiv:2607.13034：E3框架——智能体预估任务复杂度，节省91%的token消耗 https://24-ai.news/zh/news/2026-07-15/arxiv-fim-coding-agent-mid-training/ 24 AI zh-cn 2026-07-15T00:00:00Z arXiv:2607.12463：函数感知FIM中间训练将代码智能体SWE-Bench成绩提升最高+5.4分 https://24-ai.news/zh/news/2026-07-15/arxiv-pm-bench-prospective-memory-agents/ 24 AI zh-cn 2026-07-15T00:00:00Z arXiv:2607.12385：PM-Bench衡量智能体「前瞻性记忆」——最佳模型GPT-5.4仅达65.1% F1 https://24-ai.news/zh/news/2026-07-15/cncf-hami-gpu-virtualization-incubating/ 24 AI zh-cn 2026-07-15T00:00:00Z CNCF：HAMi晋升为孵化项目——面向Kubernetes AI工作负载的GPU虚拟化 https://24-ai.news/zh/news/2026-07-15/google-diffusion-model-creativity/ 24 AI zh-cn 2026-07-15T00:00:00Z Google Research：扩散模型的创造力源于score函数的「平滑化」 https://24-ai.news/zh/news/2026-07-15/langchain-agents-sandboxed-computer/ 24 AI zh-cn 2026-07-15T00:00:00Z LangChain：为什么 AI 智能体需要专属隔离计算环境（沙箱） https://24-ai.news/zh/news/2026-07-15/lfai-context-layer-docling-doclang/ 24 AI zh-cn 2026-07-15T00:00:00Z LF AI & Data：「上下文是新瓶颈」——开放基础设施 Docling 与 DocLang https://24-ai.news/zh/news/2026-07-15/openai-state-federal-ai-safety-action/ 24 AI zh-cn 2026-07-15T00:00:00Z OpenAI：倡导美国AI监管的「逆向联邦主义」 https://24-ai.news/zh/news/2026-07-15/pytorch-triton-plugin-extensions-tlx/ 24 AI zh-cn 2026-07-15T00:00:00Z PyTorch：Triton 3.7引入插件扩展与TLX——AMD MI350吞吐量提升最高+15% https://24-ai.news/zh/news/2026-07-15/vllm-tml-inkling-trillion-param-day0/ 24 AI zh-cn 2026-07-15T00:00:00Z vLLM：对TML Inkling的Day-0支持——万亿参数模型在GB200上达380 tokens/s https://24-ai.news/ja/news/2026-07-15/amd-hipvs-agentic-rag-vector-search/ 24 AI ja 2026-07-15T00:00:00Z AMD：hipVS——Instinct向けGPUベクトル検索ライブラリ、cuVS互換APIとエージェントRAGデモ付き https://24-ai.news/ja/news/2026-07-15/arxiv-cbrn-uplift-frontier-llm-eval/ 24 AI ja 2026-07-15T00:00:00Z arXiv:2607.12200：フロンティアモデルの CBRN「アップリフト」を測定する TEC フレームワーク——放射線領域のみでリスクが確認 https://24-ai.news/ja/news/2026-07-15/arxiv-e3-agents-task-complexity-estimate/ 24 AI ja 2026-07-15T00:00:00Z arXiv:2607.13034：E3フレームワーク——エージェントがタスク複雑度を事前推定し、トークンを91%削減 https://24-ai.news/ja/news/2026-07-15/arxiv-fim-coding-agent-mid-training/ 24 AI ja 2026-07-15T00:00:00Z arXiv:2607.12463：関数認識FIM中間訓練でコーディングエージェントのSWE-Benchスコアを最大+5.4向上 https://24-ai.news/ja/news/2026-07-15/arxiv-pm-bench-prospective-memory-agents/ 24 AI ja 2026-07-15T00:00:00Z arXiv:2607.12385：PM-Benchがエージェントの「展望的記憶」を測定——最高性能のGPT-5.4でもF1はわずか65.1% https://24-ai.news/ja/news/2026-07-15/cncf-hami-gpu-virtualization-incubating/ 24 AI ja 2026-07-15T00:00:00Z CNCF：HAMiがIncubatingプロジェクトに昇格——Kubernetes AIワークロード向けGPU仮想化 https://24-ai.news/ja/news/2026-07-15/google-diffusion-model-creativity/ 24 AI ja 2026-07-15T00:00:00Z Google Research：拡散モデルの創造性はscore関数の「平滑化」で説明できる https://24-ai.news/ja/news/2026-07-15/langchain-agents-sandboxed-computer/ 24 AI ja 2026-07-15T00:00:00Z LangChain：AI エージェントに専用の隔離コンピューティング環境（サンドボックス）が必要な理由 https://24-ai.news/ja/news/2026-07-15/lfai-context-layer-docling-doclang/ 24 AI ja 2026-07-15T00:00:00Z LF AI & Data：「コンテキストが新たなボトルネック」——オープン基盤 Docling と DocLang https://24-ai.news/ja/news/2026-07-15/openai-state-federal-ai-safety-action/ 24 AI ja 2026-07-15T00:00:00Z OpenAI：米国AI規制における「逆連邦主義」を提唱 https://24-ai.news/ja/news/2026-07-15/pytorch-triton-plugin-extensions-tlx/ 24 AI ja 2026-07-15T00:00:00Z PyTorch：Triton 3.7がプラグイン拡張とTLXを導入——AMD MI350でスループット最大+15% https://24-ai.news/ja/news/2026-07-15/vllm-tml-inkling-trillion-param-day0/ 24 AI ja 2026-07-15T00:00:00Z vLLM：TML InklingへのDay-0サポート——GB200で1兆パラメータモデルが380トークン/秒を達成 https://24-ai.news/ko/news/2026-07-15/amd-hipvs-agentic-rag-vector-search/ 24 AI ko 2026-07-15T00:00:00Z AMD：hipVS——Instinct용 GPU 벡터 검색 라이브러리, cuVS 호환 API 및 에이전트 RAG 데모 포함 https://24-ai.news/ko/news/2026-07-15/arxiv-cbrn-uplift-frontier-llm-eval/ 24 AI ko 2026-07-15T00:00:00Z arXiv:2607.12200：프론티어 모델의 CBRN 「업리프트」 측정을 위한 TEC 프레임워크——방사선 분야에서만 위험 확인 https://24-ai.news/ko/news/2026-07-15/arxiv-e3-agents-task-complexity-estimate/ 24 AI ko 2026-07-15T00:00:00Z arXiv:2607.13034: E3 프레임워크 — 에이전트가 태스크 복잡도를 추정하고 토큰을 91% 절감 https://24-ai.news/ko/news/2026-07-15/arxiv-fim-coding-agent-mid-training/ 24 AI ko 2026-07-15T00:00:00Z arXiv:2607.12463: 함수 인식 FIM 중간 훈련으로 코딩 에이전트 SWE-Bench 점수 최대 +5.4 향상 https://24-ai.news/ko/news/2026-07-15/arxiv-pm-bench-prospective-memory-agents/ 24 AI ko 2026-07-15T00:00:00Z arXiv:2607.12385：PM-Bench, 에이전트의 「전망적 기억」측정——최고 모델 GPT-5.4도 F1 65.1%에 그쳐 https://24-ai.news/ko/news/2026-07-15/cncf-hami-gpu-virtualization-incubating/ 24 AI ko 2026-07-15T00:00:00Z CNCF：HAMi, Incubating 프로젝트로 승격——Kubernetes AI 워크로드를 위한 GPU 가상화 https://24-ai.news/ko/news/2026-07-15/google-diffusion-model-creativity/ 24 AI ko 2026-07-15T00:00:00Z Google Research: 확산 모델의 창의성, score 함수의 「평활화」로 설명되다 https://24-ai.news/ko/news/2026-07-15/langchain-agents-sandboxed-computer/ 24 AI ko 2026-07-15T00:00:00Z LangChain：AI 에이전트에 전용 격리 컴퓨팅 환경(샌드박스)이 필요한 이유 https://24-ai.news/ko/news/2026-07-15/lfai-context-layer-docling-doclang/ 24 AI ko 2026-07-15T00:00:00Z LF AI & Data：「컨텍스트가 새로운 병목」——오픈 인프라 Docling과 DocLang https://24-ai.news/ko/news/2026-07-15/openai-state-federal-ai-safety-action/ 24 AI ko 2026-07-15T00:00:00Z OpenAI：미국 AI 규제를 위한 「역연방주의」 옹호 https://24-ai.news/ko/news/2026-07-15/pytorch-triton-plugin-extensions-tlx/ 24 AI ko 2026-07-15T00:00:00Z PyTorch：Triton 3.7, 플러그인 확장 및 TLX 도입——AMD MI350에서 처리량 최대 +15% https://24-ai.news/ko/news/2026-07-15/vllm-tml-inkling-trillion-param-day0/ 24 AI ko 2026-07-15T00:00:00Z vLLM：TML Inkling Day-0 지원——GB200에서 1조 파라미터 모델이 380 tokens/s 달성