https://24-ai.news/hr/news/2026-07-11/arxiv-causalds-data-science-agents/ 24 AI hr 2026-07-11T00:00:00Z arXiv:2607.08093: CausalDS — novi benchmark koji mjeri kauzalno zaključivanje AI agenata u data-science zadacima https://24-ai.news/hr/news/2026-07-11/arxiv-patient-centered-clinical-dialogue/ 24 AI hr 2026-07-11T00:00:00Z arXiv:2607.08625: opsežan pregled izazova kliničkih dijaloških AI sustava — od kliničkog zaključivanja do sigurnosti pacijenata https://24-ai.news/hr/news/2026-07-11/arxiv-solarchain-eval-energy-agents/ 24 AI hr 2026-07-11T00:00:00Z arXiv:2607.08681: SolarChain-Eval — fizikalno ograničen benchmark za AI agente u decentraliziranim energetskim tržištima https://24-ai.news/hr/news/2026-07-11/arxiv-tao-llm-formal-mathematics/ 24 AI hr 2026-07-11T00:00:00Z arXiv:2607.07779: Terence Tao među autorima rada o granicama velikih jezičnih modela u frontierskoj matematici https://24-ai.news/hr/news/2026-07-11/arxiv-token-flow-firewall-agent-security/ 24 AI hr 2026-07-11T00:00:00Z arXiv:2607.08395: Token-Flow Firewall — runtime nadzor koji dugoživuće AI agente štiti i spušta uspješnost napada na 12,5% https://24-ai.news/hr/news/2026-07-11/claude-code-2-1-207-auto-mode-opus48/ 24 AI hr 2026-07-11T00:00:00Z Anthropic: Claude Code v2.1.207 uključuje auto mode na svim enterprise cloudovima, mijenja zadani model na Opus 4.8 i zatvara shell-injection propust https://24-ai.news/en/news/2026-07-11/arxiv-causalds-data-science-agents/ 24 AI en 2026-07-11T00:00:00Z arXiv:2607.08093: CausalDS — a new benchmark measuring causal reasoning of AI agents in data-science tasks https://24-ai.news/en/news/2026-07-11/arxiv-patient-centered-clinical-dialogue/ 24 AI en 2026-07-11T00:00:00Z arXiv:2607.08625: comprehensive survey of challenges in clinical conversational AI systems — from clinical reasoning to patient safety https://24-ai.news/en/news/2026-07-11/arxiv-solarchain-eval-energy-agents/ 24 AI en 2026-07-11T00:00:00Z arXiv:2607.08681: SolarChain-Eval — physics-constrained benchmark for AI agents in decentralized energy markets https://24-ai.news/en/news/2026-07-11/arxiv-tao-llm-formal-mathematics/ 24 AI en 2026-07-11T00:00:00Z arXiv:2607.07779: Terence Tao among authors of paper on limits of large language models in frontier mathematics https://24-ai.news/en/news/2026-07-11/arxiv-token-flow-firewall-agent-security/ 24 AI en 2026-07-11T00:00:00Z arXiv:2607.08395: Token-Flow Firewall — runtime monitoring that protects long-running AI agents and reduces attack success rate to 12.5% https://24-ai.news/en/news/2026-07-11/claude-code-2-1-207-auto-mode-opus48/ 24 AI en 2026-07-11T00:00:00Z Anthropic: Claude Code v2.1.207 enables auto mode on all enterprise clouds, switches default model to Opus 4.8, and closes shell-injection vulnerability https://24-ai.news/de/news/2026-07-11/arxiv-causalds-data-science-agents/ 24 AI de 2026-07-11T00:00:00Z arXiv:2607.08093: CausalDS — neues Benchmark misst kausales Schlussfolgern von KI-Agenten in Data-Science-Aufgaben https://24-ai.news/de/news/2026-07-11/arxiv-patient-centered-clinical-dialogue/ 24 AI de 2026-07-11T00:00:00Z arXiv:2607.08625: umfassender Überblick über Herausforderungen klinischer konversationaler KI-Systeme — von klinischem Schlussfolgern bis zur Patientensicherheit https://24-ai.news/de/news/2026-07-11/arxiv-solarchain-eval-energy-agents/ 24 AI de 2026-07-11T00:00:00Z arXiv:2607.08681: SolarChain-Eval — physikbeschränktes Benchmark für KI-Agenten in dezentralen Energiemärkten https://24-ai.news/de/news/2026-07-11/arxiv-tao-llm-formal-mathematics/ 24 AI de 2026-07-11T00:00:00Z arXiv:2607.07779: Terence Tao als Mitautor einer Studie über die Grenzen großer Sprachmodelle in der Frontier-Mathematik https://24-ai.news/de/news/2026-07-11/arxiv-token-flow-firewall-agent-security/ 24 AI de 2026-07-11T00:00:00Z arXiv:2607.08395: Token-Flow Firewall — Runtime-Überwachung schützt langlebige KI-Agenten und senkt Angriffserfolgsrate auf 12,5 % https://24-ai.news/de/news/2026-07-11/claude-code-2-1-207-auto-mode-opus48/ 24 AI de 2026-07-11T00:00:00Z Anthropic: Claude Code v2.1.207 aktiviert Auto-Modus auf allen Enterprise-Clouds, wechselt Standardmodell zu Opus 4.8 und schließt Shell-Injection-Lücke https://24-ai.news/zh/news/2026-07-11/arxiv-causalds-data-science-agents/ 24 AI zh 2026-07-11T00:00:00Z arXiv:2607.08093：CausalDS — 衡量 AI 智能体在数据科学任务中因果推理能力的新基准 https://24-ai.news/zh/news/2026-07-11/arxiv-patient-centered-clinical-dialogue/ 24 AI zh 2026-07-11T00:00:00Z arXiv:2607.08625：临床对话 AI 系统挑战全面综述——从临床推理到患者安全 https://24-ai.news/zh/news/2026-07-11/arxiv-solarchain-eval-energy-agents/ 24 AI zh 2026-07-11T00:00:00Z arXiv:2607.08681：SolarChain-Eval — 用于去中心化能源市场 AI 智能体的物理约束基准 https://24-ai.news/zh/news/2026-07-11/arxiv-tao-llm-formal-mathematics/ 24 AI zh 2026-07-11T00:00:00Z arXiv:2607.07779：Terence Tao 参与撰写探讨大型语言模型在前沿数学中局限性的论文 https://24-ai.news/zh/news/2026-07-11/arxiv-token-flow-firewall-agent-security/ 24 AI zh 2026-07-11T00:00:00Z arXiv:2607.08395：Token-Flow Firewall — 实时监控持久 AI 智能体并将攻击成功率降至 12.5% https://24-ai.news/zh/news/2026-07-11/claude-code-2-1-207-auto-mode-opus48/ 24 AI zh 2026-07-11T00:00:00Z Anthropic: Claude Code v2.1.207 在所有企业云上启用 auto mode，将默认模型切换为 Opus 4.8 并修复 shell-injection 漏洞 https://24-ai.news/ja/news/2026-07-11/arxiv-causalds-data-science-agents/ 24 AI ja 2026-07-11T00:00:00Z arXiv:2607.08093：CausalDS — データサイエンスタスクにおける AI エージェントの因果推論を測定する新ベンチマーク https://24-ai.news/ja/news/2026-07-11/arxiv-patient-centered-clinical-dialogue/ 24 AI ja 2026-07-11T00:00:00Z arXiv:2607.08625：臨床対話 AI システムの課題に関する包括的レビュー——臨床推論から患者安全まで https://24-ai.news/ja/news/2026-07-11/arxiv-solarchain-eval-energy-agents/ 24 AI ja 2026-07-11T00:00:00Z arXiv:2607.08681：SolarChain-Eval — 分散型エネルギー市場の AI エージェント向け物理制約付きベンチマーク https://24-ai.news/ja/news/2026-07-11/arxiv-tao-llm-formal-mathematics/ 24 AI ja 2026-07-11T00:00:00Z arXiv:2607.07779：Terence Tao が共著——フロンティア数学における大規模言語モデルの限界を論じた論文 https://24-ai.news/ja/news/2026-07-11/arxiv-token-flow-firewall-agent-security/ 24 AI ja 2026-07-11T00:00:00Z arXiv:2607.08395：Token-Flow Firewall — 長期稼働 AI エージェントをリアルタイム監視し攻撃成功率を 12.5% に低下 https://24-ai.news/ja/news/2026-07-11/claude-code-2-1-207-auto-mode-opus48/ 24 AI ja 2026-07-11T00:00:00Z Anthropic: Claude Code v2.1.207 が全エンタープライズクラウドで auto mode を有効化、デフォルトモデルを Opus 4.8 に変更し shell-injection 脆弱性を修正 https://24-ai.news/ko/news/2026-07-11/arxiv-causalds-data-science-agents/ 24 AI ko 2026-07-11T00:00:00Z arXiv:2607.08093：CausalDS — 데이터 과학 작업에서 AI 에이전트의 인과 추론을 측정하는 새 벤치마크 https://24-ai.news/ko/news/2026-07-11/arxiv-patient-centered-clinical-dialogue/ 24 AI ko 2026-07-11T00:00:00Z arXiv:2607.08625：임상 대화 AI 시스템의 과제에 관한 포괄적 개요——임상 추론에서 환자 안전까지 https://24-ai.news/ko/news/2026-07-11/arxiv-solarchain-eval-energy-agents/ 24 AI ko 2026-07-11T00:00:00Z arXiv:2607.08681：SolarChain-Eval — 분산형 에너지 시장의 AI 에이전트를 위한 물리 제약 벤치마크 https://24-ai.news/ko/news/2026-07-11/arxiv-tao-llm-formal-mathematics/ 24 AI ko 2026-07-11T00:00:00Z arXiv:2607.07779：Terence Tao가 공동 저술한 최전선 수학에서 대형 언어 모델의 한계를 논한 논문 https://24-ai.news/ko/news/2026-07-11/arxiv-token-flow-firewall-agent-security/ 24 AI ko 2026-07-11T00:00:00Z arXiv:2607.08395：Token-Flow Firewall — 장기 실행 AI 에이전트를 실시간 감시하고 공격 성공률을 12.5%로 낮추다 https://24-ai.news/ko/news/2026-07-11/claude-code-2-1-207-auto-mode-opus48/ 24 AI ko 2026-07-11T00:00:00Z Anthropic: Claude Code v2.1.207이 모든 엔터프라이즈 클라우드에서 auto mode를 활성화하고 기본 모델을 Opus 4.8로 변경하며 shell-injection 취약점 수정 https://24-ai.news/hr/news/2026-07-10/amd-sglang-diffusion-rocm/ 24 AI hr 2026-07-10T00:00:00Z AMD: SGLang Diffusion na ROCm-u dovodi generiranje i uređivanje slika na Instinct GPU-e — inference framework iz LLM svijeta širi se na difuziju https://24-ai.news/hr/news/2026-07-10/arxiv-game-theory-llm-hallucination/ 24 AI hr 2026-07-10T00:00:00Z arXiv:2607.08403: teorija igara protiv halucinacija — multi-agent okvir modelira modele kao strateške igrače da smanji izmišljanje https://24-ai.news/hr/news/2026-07-10/arxiv-overthinking-reasoning-secrets/ 24 AI hr 2026-07-10T00:00:00Z arXiv:2607.08173: 'Overthinking' — pojačano zaključivanje tjera reasoning modele da odaju naučene tajne, novi ekstrakcijski napad na ICML 2026 https://24-ai.news/hr/news/2026-07-10/arxiv-super-weights-selective-training/ 24 AI hr 2026-07-10T00:00:00Z arXiv:2607.08733: 'Super težine' objašnjavaju zašto selektivni fine-tuning zakazuje — rad prihvaćen na COLM 2026 https://24-ai.news/hr/news/2026-07-10/aws-sagemaker-serverless-nemotron-finetune/ 24 AI hr 2026-07-10T00:00:00Z AWS: SageMaker donosi serverless fine-tuning NVIDIA Nemotron 3 modela s SFT, RLVR i RLAIF tehnikama https://24-ai.news/hr/news/2026-07-10/claude-code-2-1-206-release/ 24 AI hr 2026-07-10T00:00:00Z Anthropic: Claude Code v2.1.206 donosi /cd s prijedlozima putanja, /doctor savjete za CLAUDE.md i automatski git push u /commit-push-pr https://24-ai.news/hr/news/2026-07-10/github-codeql-prompt-injection-detection/ 24 AI hr 2026-07-10T00:00:00Z GitHub: CodeQL 2.26 uvodi detekciju AI prompt injectiona — prvi mainstream SAST alat koji AI napade tretira ravnopravno s klasičnima https://24-ai.news/hr/news/2026-07-10/github-copilot-code-review-prompt-fix/ 24 AI hr 2026-07-10T00:00:00Z GitHub: bolji alati pogoršali Copilot code review — prepisivanje uputa vratilo kvalitetu uz 20% niži trošak https://24-ai.news/hr/news/2026-07-10/henry-schein-dental-ai-image-verify/ 24 AI hr 2026-07-10T00:00:00Z AWS: Henry Schein One provjerava kvalitetu dentalnih rendgena AI-jem u realnom vremenu — 11 milijuna snimaka tjedno uz 1,4 sekunde latencije https://24-ai.news/hr/news/2026-07-10/langchain-openwiki-brains-agent-memory/ 24 AI hr 2026-07-10T00:00:00Z LangChain: OpenWiki Brains daje agentima proaktivnu memoriju — automatski skuplja kontekst iz Gmaila, Notiona i Gita u Markdown https://24-ai.news/hr/news/2026-07-10/openai-deutsche-telekom-ai-native-telco/ 24 AI hr 2026-07-10T00:00:00Z OpenAI: Deutsche Telekom kreće u AI-native transformaciju — partnerstvo pokriva podršku, mrežu i glasovni AI za ~245 milijuna korisnika https://24-ai.news/hr/news/2026-07-10/pytorch-free-normalization-kernel-fusion/ 24 AI hr 2026-07-10T00:00:00Z PyTorch: 'besplatna normalizacija' fuzira Layer Norm u GEMM i attention kernele — Meta cilja niži trošak treniranja velikih modela https://24-ai.news/en/news/2026-07-10/amd-sglang-diffusion-rocm/ 24 AI en 2026-07-10T00:00:00Z AMD: SGLang Diffusion on ROCm Brings Image Generation and Editing to Instinct GPUs — LLM Inference Framework Expands to Diffusion https://24-ai.news/en/news/2026-07-10/arxiv-game-theory-llm-hallucination/ 24 AI en 2026-07-10T00:00:00Z arXiv:2607.08403: Game Theory Against Hallucinations — Multi-Agent Framework Models LLMs as Strategic Players to Reduce Confabulation https://24-ai.news/en/news/2026-07-10/arxiv-overthinking-reasoning-secrets/ 24 AI en 2026-07-10T00:00:00Z arXiv:2607.08173: 'Overthinking' — Amplified Reasoning Forces Reasoning Models to Reveal Learned Secrets, New Extraction Attack at ICML 2026 https://24-ai.news/en/news/2026-07-10/arxiv-super-weights-selective-training/ 24 AI en 2026-07-10T00:00:00Z arXiv:2607.08733: 'Super Weights' Explain Why Selective Fine-Tuning Fails — Paper Accepted at COLM 2026 https://24-ai.news/en/news/2026-07-10/aws-sagemaker-serverless-nemotron-finetune/ 24 AI en 2026-07-10T00:00:00Z AWS: SageMaker Brings Serverless Fine-Tuning for NVIDIA Nemotron 3 Models with SFT, RLVR, and RLAIF Techniques https://24-ai.news/en/news/2026-07-10/claude-code-2-1-206-release/ 24 AI en 2026-07-10T00:00:00Z Anthropic: Claude Code v2.1.206 Brings /cd with Path Suggestions, /doctor Advice for CLAUDE.md, and Auto Git Push in /commit-push-pr https://24-ai.news/en/news/2026-07-10/github-codeql-prompt-injection-detection/ 24 AI en 2026-07-10T00:00:00Z GitHub: CodeQL 2.26 Introduces AI Prompt Injection Detection — First Mainstream SAST Tool to Treat AI Attacks on Par with Classic Ones https://24-ai.news/en/news/2026-07-10/github-copilot-code-review-prompt-fix/ 24 AI en 2026-07-10T00:00:00Z GitHub: Better Tools Made Copilot Code Review Worse — Rewriting Prompts Restored Quality at 20% Lower Cost https://24-ai.news/en/news/2026-07-10/henry-schein-dental-ai-image-verify/ 24 AI en 2026-07-10T00:00:00Z AWS: Henry Schein One Verifies Dental X-Ray Quality with Real-Time AI — 11 Million Images per Week at 1.4-Second Latency https://24-ai.news/en/news/2026-07-10/langchain-openwiki-brains-agent-memory/ 24 AI en 2026-07-10T00:00:00Z LangChain: OpenWiki Brains gives agents proactive memory — automatically collects context from Gmail, Notion and Git into Markdown https://24-ai.news/en/news/2026-07-10/openai-deutsche-telekom-ai-native-telco/ 24 AI en 2026-07-10T00:00:00Z OpenAI: Deutsche Telekom launches AI-native transformation — partnership covers support, network and voice AI for ~245 million customers https://24-ai.news/en/news/2026-07-10/pytorch-free-normalization-kernel-fusion/ 24 AI en 2026-07-10T00:00:00Z PyTorch: 'free normalization' fuses Layer Norm into GEMM and attention kernels — Meta targets lower training cost for large models https://24-ai.news/de/news/2026-07-10/amd-sglang-diffusion-rocm/ 24 AI de 2026-07-10T00:00:00Z AMD: SGLang Diffusion auf ROCm bringt Bildgenerierung und -bearbeitung auf Instinct-GPUs — Inferenz-Framework aus der LLM-Welt erweitert sich auf Diffusion https://24-ai.news/de/news/2026-07-10/arxiv-game-theory-llm-hallucination/ 24 AI de 2026-07-10T00:00:00Z arXiv:2607.08403: Spieltheorie gegen Halluzinationen — Multi-Agenten-Framework modelliert Sprachmodelle als strategische Spieler zur Reduktion von Fehlinformationen https://24-ai.news/de/news/2026-07-10/arxiv-overthinking-reasoning-secrets/ 24 AI de 2026-07-10T00:00:00Z arXiv:2607.08173: 'Overthinking' — Verstärktes Schlussfolgern zwingt Reasoning-Modelle zur Preisgabe gelernter Geheimnisse, neuer Extraktionsangriff auf der ICML 2026 https://24-ai.news/de/news/2026-07-10/arxiv-super-weights-selective-training/ 24 AI de 2026-07-10T00:00:00Z arXiv:2607.08733: „Super-Gewichte” erklären das Scheitern selektiven Fine-Tunings — Beitrag auf COLM 2026 angenommen https://24-ai.news/de/news/2026-07-10/aws-sagemaker-serverless-nemotron-finetune/ 24 AI de 2026-07-10T00:00:00Z AWS: SageMaker bringt serverloses Fine-Tuning für NVIDIA Nemotron 3 mit SFT, RLVR und RLAIF https://24-ai.news/de/news/2026-07-10/claude-code-2-1-206-release/ 24 AI de 2026-07-10T00:00:00Z Anthropic: Claude Code v2.1.206 bringt /cd mit Pfadvorschlägen, /doctor-Hinweise für CLAUDE.md und automatischen Git-Push in /commit-push-pr https://24-ai.news/de/news/2026-07-10/github-codeql-prompt-injection-detection/ 24 AI de 2026-07-10T00:00:00Z GitHub: CodeQL 2.26 führt KI-Prompt-Injection-Erkennung ein — erstes Mainstream-SAST-Tool, das KI-Angriffe gleichwertig mit klassischen behandelt https://24-ai.news/de/news/2026-07-10/github-copilot-code-review-prompt-fix/ 24 AI de 2026-07-10T00:00:00Z GitHub: Bessere Tools verschlechterten Copilot Code Review — Überarbeitung der Anweisungen brachte Qualität bei 20 % geringeren Kosten zurück https://24-ai.news/de/news/2026-07-10/henry-schein-dental-ai-image-verify/ 24 AI de 2026-07-10T00:00:00Z AWS: Henry Schein One prüft Qualität von Dental-Röntgenbildern in Echtzeit per KI — 11 Millionen Aufnahmen pro Woche bei 1,4 Sekunden Latenz https://24-ai.news/de/news/2026-07-10/langchain-openwiki-brains-agent-memory/ 24 AI de 2026-07-10T00:00:00Z LangChain: OpenWiki Brains gibt Agenten proaktives Gedächtnis — sammelt automatisch Kontext aus Gmail, Notion und Git in Markdown https://24-ai.news/de/news/2026-07-10/openai-deutsche-telekom-ai-native-telco/ 24 AI de 2026-07-10T00:00:00Z OpenAI: Deutsche Telekom startet KI-native Transformation — Partnerschaft umfasst Support, Netz und Sprach-KI für ~245 Millionen Kunden https://24-ai.news/de/news/2026-07-10/pytorch-free-normalization-kernel-fusion/ 24 AI de 2026-07-10T00:00:00Z PyTorch: „kostenlose Normalisierung” fusioniert Layer Norm in GEMM- und Attention-Kernel — Meta zielt auf niedrigere Trainingskosten großer Modelle https://24-ai.news/zh/news/2026-07-10/amd-sglang-diffusion-rocm/ 24 AI zh 2026-07-10T00:00:00Z AMD: ROCm上的SGLang Diffusion将图像生成与编辑带到Instinct GPU——来自LLM领域的推理框架向扩散模型扩展 https://24-ai.news/zh/news/2026-07-10/arxiv-game-theory-llm-hallucination/ 24 AI zh 2026-07-10T00:00:00Z arXiv:2607.08403：博弈论对抗幻觉——多智能体框架将模型建模为战略博弈者以减少虚构内容 https://24-ai.news/zh/news/2026-07-10/arxiv-overthinking-reasoning-secrets/ 24 AI zh 2026-07-10T00:00:00Z arXiv:2607.08173：「过度思考」——增强推理迫使reasoning模型泄露所学秘密，ICML 2026新型提取攻击 https://24-ai.news/zh/news/2026-07-10/arxiv-super-weights-selective-training/ 24 AI zh 2026-07-10T00:00:00Z arXiv:2607.08733：「超级权重」揭示选择性微调失败之谜——COLM 2026接收论文 https://24-ai.news/zh/news/2026-07-10/aws-sagemaker-serverless-nemotron-finetune/ 24 AI zh 2026-07-10T00:00:00Z AWS: SageMaker带来NVIDIA Nemotron 3模型的无服务器微调，支持SFT、RLVR和RLAIF技术 https://24-ai.news/zh/news/2026-07-10/claude-code-2-1-206-release/ 24 AI zh 2026-07-10T00:00:00Z Anthropic: Claude Code v2.1.206带来带路径建议的/cd、/doctor对CLAUDE.md的优化建议，以及/commit-push-pr中的自动git push https://24-ai.news/zh/news/2026-07-10/github-codeql-prompt-injection-detection/ 24 AI zh 2026-07-10T00:00:00Z GitHub: CodeQL 2.26引入AI提示注入检测——首款将AI攻击与经典攻击同等对待的主流SAST工具 https://24-ai.news/zh/news/2026-07-10/github-copilot-code-review-prompt-fix/ 24 AI zh 2026-07-10T00:00:00Z GitHub: 更好的工具让Copilot代码审查变差了——重写提示词在降低20%成本的同时恢复了质量 https://24-ai.news/zh/news/2026-07-10/henry-schein-dental-ai-image-verify/ 24 AI zh 2026-07-10T00:00:00Z AWS: Henry Schein One用AI实时验证牙科X光质量——每周1100万张影像，延迟1.4秒 https://24-ai.news/zh/news/2026-07-10/langchain-openwiki-brains-agent-memory/ 24 AI zh 2026-07-10T00:00:00Z LangChain: OpenWiki Brains为代理带来主动记忆——自动从Gmail、Notion和Git收集上下文并保存为Markdown https://24-ai.news/zh/news/2026-07-10/openai-deutsche-telekom-ai-native-telco/ 24 AI zh 2026-07-10T00:00:00Z OpenAI: Deutsche Telekom启动AI原生转型——合作涵盖约2.45亿用户的客服、网络和语音AI https://24-ai.news/zh/news/2026-07-10/pytorch-free-normalization-kernel-fusion/ 24 AI zh 2026-07-10T00:00:00Z PyTorch: 「免费归一化」将Layer Norm融合进GEMM和attention核——Meta瞄准降低大型模型训练成本 https://24-ai.news/ja/news/2026-07-10/amd-sglang-diffusion-rocm/ 24 AI ja 2026-07-10T00:00:00Z AMD: SGLang DiffusionをROCmで実現——LLM向けinferenceフレームワークがInstinct GPUでの画像生成・編集に拡張 https://24-ai.news/ja/news/2026-07-10/arxiv-game-theory-llm-hallucination/ 24 AI ja 2026-07-10T00:00:00Z arXiv:2607.08403: ゲーム理論でLLMの幻覚に対抗——マルチエージェント枠組みがモデルを戦略的プレイヤーとして均衡へ https://24-ai.news/ja/news/2026-07-10/arxiv-overthinking-reasoning-secrets/ 24 AI ja 2026-07-10T00:00:00Z arXiv:2607.08173: 「Overthinking」——強化された推論がreasoningモデルに学習済み秘密を暴露させる、ICML 2026採択の新たな抽出攻撃 https://24-ai.news/ja/news/2026-07-10/arxiv-super-weights-selective-training/ 24 AI ja 2026-07-10T00:00:00Z arXiv:2607.08733: 「スーパーウェイト」が選択的ファインチューニングの失敗を説明——COLM 2026採択論文 https://24-ai.news/ja/news/2026-07-10/aws-sagemaker-serverless-nemotron-finetune/ 24 AI ja 2026-07-10T00:00:00Z AWS: SageMakerがNVIDIA Nemotron 3モデルのサーバーレスファインチューニングをSFT・RLVR・RLAIFで提供開始 https://24-ai.news/ja/news/2026-07-10/claude-code-2-1-206-release/ 24 AI ja 2026-07-10T00:00:00Z Anthropic: Claude Code v2.1.206がパス提案付き/cd、CLAUDE.mdへの/doctorアドバイス、/commit-push-prの自動git pushを導入 https://24-ai.news/ja/news/2026-07-10/github-codeql-prompt-injection-detection/ 24 AI ja 2026-07-10T00:00:00Z GitHub: CodeQL 2.26がAI prompt injection検出を導入——AI攻撃をXSSやSQLインジェクションと同等に扱う初のメインストリームSASTツール https://24-ai.news/ja/news/2026-07-10/github-copilot-code-review-prompt-fix/ 24 AI ja 2026-07-10T00:00:00Z GitHub: 優れたツールがCopilot code reviewを悪化させた——プロンプトの書き直しで質を保ちつつ20%コスト削減を実現 https://24-ai.news/ja/news/2026-07-10/henry-schein-dental-ai-image-verify/ 24 AI ja 2026-07-10T00:00:00Z AWS: Henry Schein OneがAIでリアルタイムに歯科X線の品質確認——週1,100万枚の撮影を1.4秒のレイテンシで処理 https://24-ai.news/ja/news/2026-07-10/langchain-openwiki-brains-agent-memory/ 24 AI ja 2026-07-10T00:00:00Z LangChain: OpenWiki BrainsがエージェントにプロアクティブなメモリをGmail・Notion・GitからMarkdownで自動収集 https://24-ai.news/ja/news/2026-07-10/openai-deutsche-telekom-ai-native-telco/ 24 AI ja 2026-07-10T00:00:00Z OpenAI: Deutsche TelekomがAI-native変革へ——約2億4,500万ユーザーを対象にサポート・ネットワーク・音声AIをカバーするパートナーシップ https://24-ai.news/ja/news/2026-07-10/pytorch-free-normalization-kernel-fusion/ 24 AI ja 2026-07-10T00:00:00Z PyTorch: 「フリー正規化」がLayer NormをGEMMとattentionカーネルに融合——Metaが大規模モデルの学習コスト削減を目指す https://24-ai.news/ko/news/2026-07-10/amd-sglang-diffusion-rocm/ 24 AI ko 2026-07-10T00:00:00Z AMD: ROCm 기반 SGLang Diffusion이 Instinct GPU에 이미지 생성·편집 기능 제공 — LLM용 추론 프레임워크, 확산 모델로 확장 https://24-ai.news/ko/news/2026-07-10/arxiv-game-theory-llm-hallucination/ 24 AI ko 2026-07-10T00:00:00Z arXiv:2607.08403: 게임 이론으로 환각 방지 — 멀티 에이전트 프레임워크가 모델을 전략적 플레이어로 모델링해 허위 정보 생성 감소 https://24-ai.news/ko/news/2026-07-10/arxiv-overthinking-reasoning-secrets/ 24 AI ko 2026-07-10T00:00:00Z arXiv:2607.08173: 『Overthinking』 — 강화된 추론이 추론 모델로 하여금 학습된 비밀을 누설하게 만드는 ICML 2026 신규 추출 공격 https://24-ai.news/ko/news/2026-07-10/arxiv-super-weights-selective-training/ 24 AI ko 2026-07-10T00:00:00Z arXiv:2607.08733: 『슈퍼 가중치』가 선택적 파인튜닝이 실패하는 이유를 설명한다 — COLM 2026 채택 논문 https://24-ai.news/ko/news/2026-07-10/aws-sagemaker-serverless-nemotron-finetune/ 24 AI ko 2026-07-10T00:00:00Z AWS: SageMaker가 NVIDIA Nemotron 3 모델의 서버리스 파인튜닝 도입 — SFT·RLVR·RLAIF 기법 지원 https://24-ai.news/ko/news/2026-07-10/claude-code-2-1-206-release/ 24 AI ko 2026-07-10T00:00:00Z Anthropic: Claude Code v2.1.206 — 경로 제안 /cd, CLAUDE.md 조언 /doctor, /commit-push-pr의 자동 git push https://24-ai.news/ko/news/2026-07-10/github-codeql-prompt-injection-detection/ 24 AI ko 2026-07-10T00:00:00Z GitHub: CodeQL 2.26이 AI 프롬프트 인젝션 탐지 도입 — 주류 SAST 도구 최초로 AI 공격을 기존 취약점과 동등하게 처리 https://24-ai.news/ko/news/2026-07-10/github-copilot-code-review-prompt-fix/ 24 AI ko 2026-07-10T00:00:00Z GitHub: 더 나은 도구가 Copilot 코드 리뷰를 악화시켰다 — 프롬프트 재작성으로 품질 유지하면서 비용 20% 절감 https://24-ai.news/ko/news/2026-07-10/henry-schein-dental-ai-image-verify/ 24 AI ko 2026-07-10T00:00:00Z AWS: Henry Schein One, AI로 치과 엑스레이 품질을 실시간 검증 — 주당 1100만 건, 지연 시간 1.4초 https://24-ai.news/ko/news/2026-07-10/langchain-openwiki-brains-agent-memory/ 24 AI ko 2026-07-10T00:00:00Z LangChain: OpenWiki Brains가 에이전트에 능동적 메모리 부여 — Gmail·Notion·Git에서 자동으로 컨텍스트를 수집해 Markdown으로 저장 https://24-ai.news/ko/news/2026-07-10/openai-deutsche-telekom-ai-native-telco/ 24 AI ko 2026-07-10T00:00:00Z OpenAI: Deutsche Telekom, AI 네이티브 전환 착수 — 약 2억 4500만 고객의 지원·네트워크·음성 AI 포괄하는 파트너십 https://24-ai.news/ko/news/2026-07-10/pytorch-free-normalization-kernel-fusion/ 24 AI ko 2026-07-10T00:00:00Z PyTorch: 『무료 정규화』로 Layer Norm을 GEMM·어텐션 커널에 융합 — Meta, 대규모 모델 학습 비용 절감 목표 https://24-ai.news/hr/news/2026-07-09/amd-flydsl-python-gpu-kernels/ 24 AI hr 2026-07-09T00:00:00Z AMD: FlyDSL — Python DSL koji obećava performanse ručno pisanih HIP C++ GPU kernela uz znatno manje koda https://24-ai.news/hr/news/2026-07-09/anthropic-bernanke-benefit-trust/ 24 AI hr 2026-07-09T00:00:00Z Anthropic: nobelovac Ben Bernanke, bivši šef Fed-a, ulazi u Long-Term Benefit Trust — tijelo koje nadzire misiju kompanije https://24-ai.news/hr/news/2026-07-09/anthropic-hard-questions-initiative/ 24 AI hr 2026-07-09T00:00:00Z Anthropic: inicijativa 'Hard Questions' poziva javnost da postavlja najteža pitanja o AI-ju — uz obvezu javnog izvještavanja https://24-ai.news/hr/news/2026-07-09/anthropic-reflect-usage-dashboard/ 24 AI hr 2026-07-09T00:00:00Z Anthropic: Reflect — dashboard koji pokazuje kako koristiš Claude, s tihim satima i pitanjima za samorefleksiju https://24-ai.news/hr/news/2026-07-09/arxiv-harness-effect-token-economics/ 24 AI hr 2026-07-09T00:00:00Z arXiv:2607.06906: 'Harness Effect' — dizajn orkestracije reže trošak AI agenta 41%, više nego promjena modela https://24-ai.news/hr/news/2026-07-09/github-copilot-repository-overview/ 24 AI hr 2026-07-09T00:00:00Z GitHub: Copilot sada nudi AI pregled repozitorija — automatski sažetak svrhe, tehnologija i načina doprinošenja https://24-ai.news/hr/news/2026-07-09/google-sensorfm-wearable-health/ 24 AI hr 2026-07-09T00:00:00Z Google: SensorFM — foundation model treniran na bilijun minuta wearable podataka pobjeđuje na 34 od 35 zdravstvenih zadataka https://24-ai.news/hr/news/2026-07-09/ibm-bob-multi-agent-mainframe/ 24 AI hr 2026-07-09T00:00:00Z IBM: Bob postaje multi-agent platforma — COBOL modernizacija devetomjesečnog projekta završena u 3 dana https://24-ai.news/hr/news/2026-07-09/meta-muse-spark-1-1-reasoning/ 24 AI hr 2026-07-09T00:00:00Z Meta: Muse Spark 1.1 donosi multimodalno zaključivanje s milijun tokena konteksta i koordinaciju subagenata kroz Model API https://24-ai.news/hr/news/2026-07-09/microsoft-aurora-1-5-weather-sota/ 24 AI hr 2026-07-09T00:00:00Z Microsoft: Aurora 1.5 open-source model nadmašuje ECMWF ensemble na 88,9% varijabli — novi standard AI prognoze vremena https://24-ai.news/hr/news/2026-07-09/mistral-studio-prompt-skill-governance/ 24 AI hr 2026-07-09T00:00:00Z Mistral: Studio dobiva version control za promptove i skillove — nepromjenjive verzije, audit tragovi i rollback https://24-ai.news/hr/news/2026-07-09/ollama-88m-funding-local-ai/ 24 AI hr 2026-07-09T00:00:00Z Ollama: 88 milijuna dolara financiranja za lokalni AI — 8,9 milijuna developera i Docker osnivač među investitorima https://24-ai.news/hr/news/2026-07-09/openai-bio-bug-bounty-gpt-5-5/ 24 AI hr 2026-07-09T00:00:00Z OpenAI: prvi Bio Bug Bounty — istraživači pozvani da pronađu biosigurnosne propuste u GPT-5.5 modelu https://24-ai.news/hr/news/2026-07-09/openai-chatgpt-work-agent/ 24 AI hr 2026-07-09T00:00:00Z OpenAI: ChatGPT postaje agent za ozbiljan posao — radi satima kroz aplikacije i datoteke do gotovog rezultata https://24-ai.news/hr/news/2026-07-09/openai-gpt-5-6-sol-terra-luna/ 24 AI hr 2026-07-09T00:00:00Z OpenAI: GPT-5.6 stiže u tri varijante — Sol, Terra i Luna, s multi-agent orkestracijom i istim danom u GitHub Copilotu https://24-ai.news/hr/news/2026-07-09/xai-grok-4-5-coding-agentic/ 24 AI hr 2026-07-09T00:00:00Z xAI: Grok 4.5 — novi flagship za kodiranje i agentski rad po cijeni od 2 dolara za milijun ulaznih tokena https://24-ai.news/en/news/2026-07-09/amd-flydsl-python-gpu-kernels/ 24 AI en 2026-07-09T00:00:00Z AMD: FlyDSL — Python DSL that promises performance of hand-written HIP C++ GPU kernels with far less code https://24-ai.news/en/news/2026-07-09/anthropic-bernanke-benefit-trust/ 24 AI en 2026-07-09T00:00:00Z Anthropic: Nobel laureate Ben Bernanke, former Fed chair, joins the Long-Term Benefit Trust — the body overseeing the company's mission https://24-ai.news/en/news/2026-07-09/anthropic-hard-questions-initiative/ 24 AI en 2026-07-09T00:00:00Z Anthropic: 'Hard Questions' initiative invites the public to ask the toughest questions about AI — with a commitment to public reporting https://24-ai.news/en/news/2026-07-09/anthropic-reflect-usage-dashboard/ 24 AI en 2026-07-09T00:00:00Z Anthropic: Reflect — dashboard showing how you use Claude, with quiet hours and self-reflection prompts https://24-ai.news/en/news/2026-07-09/arxiv-harness-effect-token-economics/ 24 AI en 2026-07-09T00:00:00Z arXiv:2607.06906: 'Harness Effect' — orchestration design cuts AI agent cost by 41%, more than changing the model https://24-ai.news/en/news/2026-07-09/github-copilot-repository-overview/ 24 AI en 2026-07-09T00:00:00Z GitHub: Copilot now offers AI repository overview — automatic summary of purpose, technologies, and how to contribute https://24-ai.news/en/news/2026-07-09/google-sensorfm-wearable-health/ 24 AI en 2026-07-09T00:00:00Z Google: SensorFM — foundation model trained on one trillion minutes of wearable data wins on 34 of 35 health tasks https://24-ai.news/en/news/2026-07-09/ibm-bob-multi-agent-mainframe/ 24 AI en 2026-07-09T00:00:00Z IBM: Bob becomes a multi-agent platform — nine-month COBOL modernization project completed in 3 days https://24-ai.news/en/news/2026-07-09/meta-muse-spark-1-1-reasoning/ 24 AI en 2026-07-09T00:00:00Z Meta: Muse Spark 1.1 brings multimodal reasoning with one million token context and sub-agent coordination via Model API https://24-ai.news/en/news/2026-07-09/microsoft-aurora-1-5-weather-sota/ 24 AI en 2026-07-09T00:00:00Z Microsoft: Aurora 1.5 open-source model outperforms ECMWF ensemble on 88.9% of variables — new standard in AI weather forecasting https://24-ai.news/en/news/2026-07-09/mistral-studio-prompt-skill-governance/ 24 AI en 2026-07-09T00:00:00Z Mistral: Studio gets version control for prompts and skills — immutable versions, audit trails, and rollback https://24-ai.news/en/news/2026-07-09/ollama-88m-funding-local-ai/ 24 AI en 2026-07-09T00:00:00Z Ollama: $88 million in funding for local AI — 8.9 million developers and Docker's founder among investors https://24-ai.news/en/news/2026-07-09/openai-bio-bug-bounty-gpt-5-5/ 24 AI en 2026-07-09T00:00:00Z OpenAI: first Bio Bug Bounty — researchers invited to find biosecurity vulnerabilities in the GPT-5.5 model https://24-ai.news/en/news/2026-07-09/openai-chatgpt-work-agent/ 24 AI en 2026-07-09T00:00:00Z OpenAI: ChatGPT becomes an agent for serious work — operates for hours through apps and files until the job is done https://24-ai.news/en/news/2026-07-09/openai-gpt-5-6-sol-terra-luna/ 24 AI en 2026-07-09T00:00:00Z OpenAI: GPT-5.6 arrives in three variants — Sol, Terra and Luna, with multi-agent orchestration and same-day availability in GitHub Copilot https://24-ai.news/en/news/2026-07-09/xai-grok-4-5-coding-agentic/ 24 AI en 2026-07-09T00:00:00Z xAI: Grok 4.5 — new flagship for coding and agentic work at $2 per million input tokens https://24-ai.news/de/news/2026-07-09/amd-flydsl-python-gpu-kernels/ 24 AI de 2026-07-09T00:00:00Z AMD: FlyDSL — Python-DSL verspricht Leistung manuell geschriebener HIP-C++-GPU-Kernel mit deutlich weniger Code https://24-ai.news/de/news/2026-07-09/anthropic-bernanke-benefit-trust/ 24 AI de 2026-07-09T00:00:00Z Anthropic: Nobelpreisträger Ben Bernanke, ehemaliger Fed-Chef, tritt dem Long-Term Benefit Trust bei — dem Gremium zur Überwachung der Unternehmensmission https://24-ai.news/de/news/2026-07-09/anthropic-hard-questions-initiative/ 24 AI de 2026-07-09T00:00:00Z Anthropic: Initiative 'Hard Questions' lädt die Öffentlichkeit ein, die schwierigsten KI-Fragen zu stellen — mit Berichtspflicht https://24-ai.news/de/news/2026-07-09/anthropic-reflect-usage-dashboard/ 24 AI de 2026-07-09T00:00:00Z Anthropic: Reflect — Dashboard zeigt, wie Sie Claude nutzen, mit Ruhezeiten und Selbstreflexionsfragen https://24-ai.news/de/news/2026-07-09/arxiv-harness-effect-token-economics/ 24 AI de 2026-07-09T00:00:00Z arXiv:2607.06906: 'Harness Effect' — Orchestrierungsdesign senkt KI-Agenten-Kosten um 41 % — mehr als ein Modellwechsel https://24-ai.news/de/news/2026-07-09/github-copilot-repository-overview/ 24 AI de 2026-07-09T00:00:00Z GitHub: Copilot bietet jetzt KI-Repository-Übersicht — automatische Zusammenfassung von Zweck, Technologien und Beitragshinweisen https://24-ai.news/de/news/2026-07-09/google-sensorfm-wearable-health/ 24 AI de 2026-07-09T00:00:00Z Google: SensorFM — Foundation-Modell auf einer Billion Minuten Wearable-Daten gewinnt 34 von 35 Gesundheitsaufgaben https://24-ai.news/de/news/2026-07-09/ibm-bob-multi-agent-mainframe/ 24 AI de 2026-07-09T00:00:00Z IBM: Bob wird zur Multi-Agent-Plattform — neunmonatiges COBOL-Modernisierungsprojekt in 3 Tagen abgeschlossen https://24-ai.news/de/news/2026-07-09/meta-muse-spark-1-1-reasoning/ 24 AI de 2026-07-09T00:00:00Z Meta: Muse Spark 1.1 bringt multimodales Reasoning mit einer Million Token Kontext und Sub-Agenten-Koordination über Model API https://24-ai.news/de/news/2026-07-09/microsoft-aurora-1-5-weather-sota/ 24 AI de 2026-07-09T00:00:00Z Microsoft: Open-Source-Modell Aurora 1.5 übertrifft ECMWF-Ensemble bei 88,9 % der Variablen — neuer Standard in der KI-Wettervorhersage https://24-ai.news/de/news/2026-07-09/mistral-studio-prompt-skill-governance/ 24 AI de 2026-07-09T00:00:00Z Mistral: Studio erhält Versionskontrolle für Prompts und Skills — unveränderliche Versionen, Audit-Protokolle und Rollback https://24-ai.news/de/news/2026-07-09/ollama-88m-funding-local-ai/ 24 AI de 2026-07-09T00:00:00Z Ollama: 88 Millionen Dollar Finanzierung für lokale KI — 8,9 Millionen Entwickler und Docker-Gründer als Investoren https://24-ai.news/de/news/2026-07-09/openai-bio-bug-bounty-gpt-5-5/ 24 AI de 2026-07-09T00:00:00Z OpenAI: erstes Bio Bug Bounty — Forscher eingeladen, Biosicherheitslücken im GPT-5.5-Modell zu finden https://24-ai.news/de/news/2026-07-09/openai-chatgpt-work-agent/ 24 AI de 2026-07-09T00:00:00Z OpenAI: ChatGPT wird zum Agenten für ernsthafte Arbeit — stundenlang durch Apps und Dateien bis zum fertigen Ergebnis https://24-ai.news/de/news/2026-07-09/openai-gpt-5-6-sol-terra-luna/ 24 AI de 2026-07-09T00:00:00Z OpenAI: GPT-5.6 in drei Varianten — Sol, Terra und Luna mit Multi-Agent-Orchestrierung und sofortiger GitHub-Copilot-Integration https://24-ai.news/de/news/2026-07-09/xai-grok-4-5-coding-agentic/ 24 AI de 2026-07-09T00:00:00Z xAI: Grok 4.5 — neues Flagship für Coding und agentische Arbeit für 2 Dollar pro Million Eingabe-Token https://24-ai.news/zh/news/2026-07-09/amd-flydsl-python-gpu-kernels/ 24 AI zh 2026-07-09T00:00:00Z AMD: FlyDSL——Python DSL 承诺以更少代码实现手写 HIP C++ GPU 核函数的性能 https://24-ai.news/zh/news/2026-07-09/anthropic-bernanke-benefit-trust/ 24 AI zh 2026-07-09T00:00:00Z Anthropic: 诺贝尔奖得主、前美联储主席 Ben Bernanke 加入 Long-Term Benefit Trust——公司使命监督机构 https://24-ai.news/zh/news/2026-07-09/anthropic-hard-questions-initiative/ 24 AI zh 2026-07-09T00:00:00Z Anthropic: 「Hard Questions」倡议邀请公众提出最难的 AI 问题——并承诺公开报告 https://24-ai.news/zh/news/2026-07-09/anthropic-reflect-usage-dashboard/ 24 AI zh 2026-07-09T00:00:00Z Anthropic: Reflect——展示你如何使用 Claude 的控制台，含免打扰时段和自我反思问题 https://24-ai.news/zh/news/2026-07-09/arxiv-harness-effect-token-economics/ 24 AI zh 2026-07-09T00:00:00Z arXiv:2607.06906：「Harness Effect」——编排设计将 AI 智能体成本削减 41%，效果超过更换模型 https://24-ai.news/zh/news/2026-07-09/github-copilot-repository-overview/ 24 AI zh 2026-07-09T00:00:00Z GitHub: Copilot 现在提供 AI 仓库概览——自动总结目的、技术栈和贡献方式 https://24-ai.news/zh/news/2026-07-09/google-sensorfm-wearable-health/ 24 AI zh 2026-07-09T00:00:00Z Google: SensorFM——基于万亿分钟可穿戴数据训练的基础模型，在 35 项健康任务中赢得 34 项 https://24-ai.news/zh/news/2026-07-09/ibm-bob-multi-agent-mainframe/ 24 AI zh 2026-07-09T00:00:00Z IBM: Bob 成为多智能体平台——COBOL 现代化九个月的项目在 3 天内完成 https://24-ai.news/zh/news/2026-07-09/meta-muse-spark-1-1-reasoning/ 24 AI zh 2026-07-09T00:00:00Z Meta: Muse Spark 1.1 带来百万 token 上下文的多模态推理，并通过 Model API 协调子智能体 https://24-ai.news/zh/news/2026-07-09/microsoft-aurora-1-5-weather-sota/ 24 AI zh 2026-07-09T00:00:00Z Microsoft: 开源模型 Aurora 1.5 在 88.9% 的变量上超越 ECMWF 集合预报——AI 天气预报新标准 https://24-ai.news/zh/news/2026-07-09/mistral-studio-prompt-skill-governance/ 24 AI zh 2026-07-09T00:00:00Z Mistral: Studio 新增提示词和技能的版本控制——不可变版本、审计追踪和回滚 https://24-ai.news/zh/news/2026-07-09/ollama-88m-funding-local-ai/ 24 AI zh 2026-07-09T00:00:00Z Ollama: 获得 8800 万美元融资用于本地 AI——890 万开发者，Docker 创始人参与投资 https://24-ai.news/zh/news/2026-07-09/openai-bio-bug-bounty-gpt-5-5/ 24 AI zh 2026-07-09T00:00:00Z OpenAI: 首个生物漏洞赏金计划——邀请研究人员发现 GPT-5.5 的生物安全缺陷 https://24-ai.news/zh/news/2026-07-09/openai-chatgpt-work-agent/ 24 AI zh 2026-07-09T00:00:00Z OpenAI: ChatGPT 成为真正的工作智能体——可跨应用操作，自主工作数小时直至完成 https://24-ai.news/zh/news/2026-07-09/openai-gpt-5-6-sol-terra-luna/ 24 AI zh 2026-07-09T00:00:00Z OpenAI: GPT-5.6 推出三款变体——Sol、Terra 与 Luna，内置多智能体编排，GitHub Copilot 同日上线 https://24-ai.news/zh/news/2026-07-09/xai-grok-4-5-coding-agentic/ 24 AI zh 2026-07-09T00:00:00Z xAI: Grok 4.5——面向编程和智能体任务的新旗舰，每百万输入 token 仅需 2 美元 https://24-ai.news/ja/news/2026-07-09/amd-flydsl-python-gpu-kernels/ 24 AI ja 2026-07-09T00:00:00Z AMD: FlyDSL——より少ないコードで手書き HIP C++ GPU カーネルの性能を約束する Python DSL https://24-ai.news/ja/news/2026-07-09/anthropic-bernanke-benefit-trust/ 24 AI ja 2026-07-09T00:00:00Z Anthropic: ノーベル賞受賞者・元 FRB 議長 Ben Bernanke が Long-Term Benefit Trust に就任——企業使命の監督機関 https://24-ai.news/ja/news/2026-07-09/anthropic-hard-questions-initiative/ 24 AI ja 2026-07-09T00:00:00Z Anthropic: 「Hard Questions」イニシアチブが AI に関する最も難しい質問を公募——公開報告の義務を伴う https://24-ai.news/ja/news/2026-07-09/anthropic-reflect-usage-dashboard/ 24 AI ja 2026-07-09T00:00:00Z Anthropic: Reflect——Claude の使い方を示すダッシュボード、サイレント時間帯と内省的な質問つき https://24-ai.news/ja/news/2026-07-09/arxiv-harness-effect-token-economics/ 24 AI ja 2026-07-09T00:00:00Z arXiv:2607.06906：「Harness Effect」——オーケストレーション設計が AI エージェントのコストを 41% 削減、モデル変更より効果的 https://24-ai.news/ja/news/2026-07-09/github-copilot-repository-overview/ 24 AI ja 2026-07-09T00:00:00Z GitHub: Copilot がリポジトリの AI 概要を提供——目的・技術スタック・貢献方法を自動生成 https://24-ai.news/ja/news/2026-07-09/google-sensorfm-wearable-health/ 24 AI ja 2026-07-09T00:00:00Z Google: SensorFM——1 兆分超えのウェアラブルデータで学習した基盤モデル、35 健康タスク中 34 で勝利 https://24-ai.news/ja/news/2026-07-09/ibm-bob-multi-agent-mainframe/ 24 AI ja 2026-07-09T00:00:00Z IBM: Bob がマルチエージェントプラットフォームに——COBOL 現代化の 9 か月プロジェクトが 3 日で完了 https://24-ai.news/ja/news/2026-07-09/meta-muse-spark-1-1-reasoning/ 24 AI ja 2026-07-09T00:00:00Z Meta: Muse Spark 1.1 が 100 万トークンコンテキストのマルチモーダル推論と Model API 経由のサブエージェント調整を実現 https://24-ai.news/ja/news/2026-07-09/microsoft-aurora-1-5-weather-sota/ 24 AI ja 2026-07-09T00:00:00Z Microsoft: オープンソースモデル Aurora 1.5 が 88.9% の変数で ECMWF アンサンブルを超越——AI 天気予報の新基準 https://24-ai.news/ja/news/2026-07-09/mistral-studio-prompt-skill-governance/ 24 AI ja 2026-07-09T00:00:00Z Mistral: Studio にプロンプトとスキルのバージョン管理が登場——不変バージョン、監査証跡、ロールバック https://24-ai.news/ja/news/2026-07-09/ollama-88m-funding-local-ai/ 24 AI ja 2026-07-09T00:00:00Z Ollama: ローカル AI 向けに 8,800 万ドルの資金調達——890 万人の開発者、Docker 創設者も投資家に https://24-ai.news/ja/news/2026-07-09/openai-bio-bug-bounty-gpt-5-5/ 24 AI ja 2026-07-09T00:00:00Z OpenAI: 初の Bio Bug Bounty——GPT-5.5 の生物安全上の脆弱性を発見する研究者を募集 https://24-ai.news/ja/news/2026-07-09/openai-chatgpt-work-agent/ 24 AI ja 2026-07-09T00:00:00Z OpenAI: ChatGPT が本格的な仕事のエージェントに——アプリをまたいで操作し、何時間も自律的に作業して成果を出す https://24-ai.news/ja/news/2026-07-09/openai-gpt-5-6-sol-terra-luna/ 24 AI ja 2026-07-09T00:00:00Z OpenAI: GPT-5.6 が Sol・Terra・Luna の3バリアントで登場——マルチエージェント編成と GitHub Copilot への初日対応 https://24-ai.news/ja/news/2026-07-09/xai-grok-4-5-coding-agentic/ 24 AI ja 2026-07-09T00:00:00Z xAI: Grok 4.5——コーディングとエージェントタスク向け新フラッグシップ、100 万入力トークンわずか 2 ドル https://24-ai.news/ko/news/2026-07-09/amd-flydsl-python-gpu-kernels/ 24 AI ko 2026-07-09T00:00:00Z AMD: FlyDSL——더 적은 코드로 수작업 HIP C++ GPU 커널의 성능을 약속하는 Python DSL https://24-ai.news/ko/news/2026-07-09/anthropic-bernanke-benefit-trust/ 24 AI ko 2026-07-09T00:00:00Z Anthropic: 노벨상 수상자·전 FRB 의장 Ben Bernanke, Long-Term Benefit Trust 합류——회사 사명 감독 기구 https://24-ai.news/ko/news/2026-07-09/anthropic-hard-questions-initiative/ 24 AI ko 2026-07-09T00:00:00Z Anthropic: 「Hard Questions」 이니셔티브, AI에 관한 가장 어려운 질문을 공개 모집——공개 보고 의무 포함 https://24-ai.news/ko/news/2026-07-09/anthropic-reflect-usage-dashboard/ 24 AI ko 2026-07-09T00:00:00Z Anthropic: Reflect——Claude 사용 방식을 보여주는 대시보드, 방해 금지 시간대와 자기 성찰 질문 포함 https://24-ai.news/ko/news/2026-07-09/arxiv-harness-effect-token-economics/ 24 AI ko 2026-07-09T00:00:00Z arXiv:2607.06906: 「Harness Effect」——오케스트레이션 설계가 AI 에이전트 비용을 41% 절감, 모델 교체보다 효과적 https://24-ai.news/ko/news/2026-07-09/github-copilot-repository-overview/ 24 AI ko 2026-07-09T00:00:00Z GitHub: Copilot이 이제 AI 저장소 개요 제공——목적, 기술 스택, 기여 방법을 자동으로 요약 https://24-ai.news/ko/news/2026-07-09/google-sensorfm-wearable-health/ 24 AI ko 2026-07-09T00:00:00Z Google: SensorFM——1조 분 이상 웨어러블 데이터로 학습한 기반 모델, 35개 건강 과제 중 34개 승리 https://24-ai.news/ko/news/2026-07-09/ibm-bob-multi-agent-mainframe/ 24 AI ko 2026-07-09T00:00:00Z IBM: Bob, 멀티 에이전트 플랫폼으로 진화——COBOL 현대화 9개월 프로젝트를 3일 만에 완료 https://24-ai.news/ko/news/2026-07-09/meta-muse-spark-1-1-reasoning/ 24 AI ko 2026-07-09T00:00:00Z Meta: Muse Spark 1.1, 100만 토큰 컨텍스트 멀티모달 추론과 Model API를 통한 서브 에이전트 조정 제공 https://24-ai.news/ko/news/2026-07-09/microsoft-aurora-1-5-weather-sota/ 24 AI ko 2026-07-09T00:00:00Z Microsoft: 오픈소스 모델 Aurora 1.5, 88.9% 변수에서 ECMWF 앙상블 초과——AI 기상 예보의 새 기준 https://24-ai.news/ko/news/2026-07-09/mistral-studio-prompt-skill-governance/ 24 AI ko 2026-07-09T00:00:00Z Mistral: Studio에 프롬프트와 스킬 버전 관리 도입——불변 버전, 감사 추적, 롤백 https://24-ai.news/ko/news/2026-07-09/ollama-88m-funding-local-ai/ 24 AI ko 2026-07-09T00:00:00Z Ollama: 로컬 AI를 위한 8800만 달러 투자 유치——890만 개발자, Docker 창업자도 투자자로 참여 https://24-ai.news/ko/news/2026-07-09/openai-bio-bug-bounty-gpt-5-5/ 24 AI ko 2026-07-09T00:00:00Z OpenAI: 첫 번째 Bio Bug Bounty——GPT-5.5의 생물 안전 취약점을 찾을 연구자 모집 https://24-ai.news/ko/news/2026-07-09/openai-chatgpt-work-agent/ 24 AI ko 2026-07-09T00:00:00Z OpenAI: ChatGPT, 실제 업무 에이전트로 진화——앱 전반에 걸쳐 작업하고 수 시간 동안 자율적으로 일해 결과물 완성 https://24-ai.news/ko/news/2026-07-09/openai-gpt-5-6-sol-terra-luna/ 24 AI ko 2026-07-09T00:00:00Z OpenAI: GPT-5.6, Sol·Terra·Luna 세 가지 변형으로 출시——멀티 에이전트 오케스트레이션과 GitHub Copilot 당일 지원 https://24-ai.news/ko/news/2026-07-09/xai-grok-4-5-coding-agentic/ 24 AI ko 2026-07-09T00:00:00Z xAI: Grok 4.5——코딩 및 에이전트 작업을 위한 새 플래그십, 백만 입력 토큰당 단 2달러