<?xml version="1.0" encoding="UTF-8"?>
<urlset xmlns="http://www.sitemaps.org/schemas/sitemap/0.9"
        xmlns:news="http://www.google.com/schemas/sitemap-news/0.9">
  <url>
    <loc>https://www.datacubeai.space/de/news/2026-06-13/tech-1818</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>de</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Anthropic hat ein KI-Modell veröffentlicht, das zuvor als zu gefährlich galt. Ein Entwickler ließ es in einem...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/de/news/2026-06-13/tech-1819</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>de</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Paca ist eine Open-Source-Alternative zu Jira, die speziell für die Zusammenarbeit von Menschen und KI-Agenten...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/de/news/2026-06-13/tech-1821</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>de</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Ein eindringliches Plädoyer für Open-Source-KI: Wenn Intelligenz nur von wenigen Konzernen gemietet werden...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/de/news/2026-06-13/tech-1822</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>de</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>„Count Anything“ ist das erste KI-Modell, das Objekte in beliebigen Bildern zuverlässig zählen kann – von...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/de/news/2026-06-13/tech-1823</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>de</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Microsoft-CEO Satya Nadella warnt vor „Token-Maxing“ – dem unnötigen Einsatz der teuersten KI-Modelle für...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/de/news/2026-06-13/tech-1824</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>de</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Microsoft-Forscher haben SkillOpt entwickelt – eine Methode, die GPT-5.5 mit einer einzigen trainierten...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/de/news/2026-06-13/tech-1826</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>de</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Anthropics Claude Fable 5 erreicht 88% Genauigkeit bei den schwersten Mathe-Aufgaben – ein riesiger Sprung von...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/de/news/2026-06-13/tech-1827</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>de</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Meta stellt sich auf milliardenschwere KI-Kosten ein: Ein internes Memo kündigt ab 2027 strikte Budgets und...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/de/news/2026-06-13/tech-1828</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>de</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Eine Anthropic-Umfrage unter 52.000 Amerikanern zeigt: 64% befürchten Jobverluste durch KI, 56% haben Angst...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/de/news/2026-06-13/tech-1829</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>de</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>OpenAI startet einen Preiskampf: Nutzer des Codex-Kodieragenten können nicht genutzte Ratenbegrenzungen jetzt...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/en/news/2026-06-13/tech-1818</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>en</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Anthropic released a model previously considered too dangerous. A developer used it to build a complete game...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/en/news/2026-06-13/tech-1819</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>en</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Paca is an open-source Jira alternative designed for collaboration between humans and AI agents. It treats AI...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/en/news/2026-06-13/tech-1821</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>en</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>A passionate argument for open-source AI: if intelligence can only be rented from a few closed institutions...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/en/news/2026-06-13/tech-1822</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>en</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>&apos;Count Anything&apos; is the first AI model that can reliably count objects in any image – from crowds to cells...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/en/news/2026-06-13/tech-1823</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>en</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Microsoft CEO Satya Nadella warns against &apos;token-maxing&apos; – using the most expensive AI models for trivial...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/en/news/2026-06-13/tech-1824</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>en</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Microsoft researchers developed SkillOpt, a method that boosts GPT-5.5 using just a trained Markdown file. It...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/en/news/2026-06-13/tech-1826</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>en</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Anthropic&apos;s Claude Fable 5 achieves 88% accuracy on the hardest math problems – a massive leap from below 10%...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/en/news/2026-06-13/tech-1827</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>en</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Meta braces for billions in AI costs: an internal memo announces strict budgets and a central &apos;AI Gateway&apos;...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/en/news/2026-06-13/tech-1828</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>en</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>An Anthropic survey of 52,000 Americans reveals: 64% fear job losses from AI, 56% worry about losing...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/en/news/2026-06-13/tech-1829</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>en</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>OpenAI kicks off a price war by letting Codex coding agent users bank and manually activate unused rate-limit...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/zh/news/2026-06-13/tech-1818</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>zh</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Anthropic 发布了一款之前被认为过于危险的模型。一位开发者使用它在一次尝试中构建了一个完整的游戏——在进行了 45 分钟的推理并花费超过 €20 的 token 成本后，它生成了一个包含...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/zh/news/2026-06-13/tech-1819</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>zh</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Paca 是一个专为人类与 AI 代理协作而设计的开源 Jira 替代品。它将 AI 视为 Scrum 工作流中的平等团队成员，而不仅仅是一个聊天机器人插件。</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/zh/news/2026-06-13/tech-1821</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>zh</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>关于开源 AI 的热情论点：如果智能只能从少数几家封闭机构租赁，我们就会失去运营自由。免费的 AI 对教育、科学和公共服务具有存在意义。</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/zh/news/2026-06-13/tech-1822</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>zh</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>&apos;Count Anything&apos; 是第一个能够可靠地计算任何图像中对象的 AI 模型——从人群到细胞。这项任务听起来简单，但对 AI 来说却出人意料地困难。</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/zh/news/2026-06-13/tech-1823</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>zh</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>微软首席执行官 Satya Nadella 警告不要进行‘token-maxing’——即用最昂贵的 AI 模型处理琐碎任务。他承认自己也对强大的模型上瘾。生产力的提升必须能够证明 token 成本的合理性。</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/zh/news/2026-06-13/tech-1824</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>zh</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>微软研究人员开发了 SkillOpt，这是一种仅使用一个经过训练的 Markdown 文件就能提升 GPT-5.5 的方法。它在不进行昂贵的微调的情况下，优化了 AI 代理的指令。</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/zh/news/2026-06-13/tech-1826</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>zh</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Anthropic 的 Claude Fable 5 在最难的数学问题上达到 88% 的准确率——相比 2026 年初低于 10% 的水平，这是一次巨大的飞跃。GPT-5.5 的准确率约为 75%。AI...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/zh/news/2026-06-13/tech-1827</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>zh</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Meta 正在为数十亿美元的 AI 成本作准备：一份内部备忘录宣布从 2027 年起实施严格预算并推出中心化的 &apos;AI Gateway&apos; 仪表板。CTO 警告：‘每一步行动都会产生费用。’</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/zh/news/2026-06-13/tech-1828</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>zh</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Anthropic 对 52,000 名美国人的调查显示：64% 担心 AI 导致失业，56% 担心失去独立思考能力。日常使用 AI 的用户担忧较少。大多数人反对在个人决策中使用 AI。</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/zh/news/2026-06-13/tech-1829</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>zh</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>OpenAI 通过允许 Codex 编码代理用户存储并手动激活未使用的速率限制重置，拉开了价格战的序幕。这降低了重度 AI 使用的成本，并提高了灵活性。</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/fr/news/2026-06-13/tech-1818</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>fr</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Anthropic a publié un modèle précédemment considéré comme trop dangereux. Un développeur l&apos;a utilisé pour...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/fr/news/2026-06-13/tech-1819</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>fr</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Paca est une alternative open source à Jira conçue pour la collaboration entre humains et agents d&apos;IA. Elle...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/fr/news/2026-06-13/tech-1821</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>fr</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Un argument passionné pour l&apos;IA open source : si l&apos;intelligence ne peut être louée qu&apos;auprès de quelques...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/fr/news/2026-06-13/tech-1822</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>fr</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>&apos;Count Anything&apos; est le premier modèle d&apos;IA capable de compter de manière fiable des objets dans n&apos;importe...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/fr/news/2026-06-13/tech-1823</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>fr</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Le PDG de Microsoft, Satya Nadella, met en garde contre le &apos;maximisation de tokens&apos; – utiliser les modèles...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/fr/news/2026-06-13/tech-1824</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>fr</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Des chercheurs de Microsoft ont développé SkillOpt, une méthode qui améliore GPT-5.5 en utilisant simplement...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/fr/news/2026-06-13/tech-1826</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>fr</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Claude Fable 5 d&apos;Anthropic atteint une précision de 88 % sur les problèmes de mathématiques les plus...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/fr/news/2026-06-13/tech-1827</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>fr</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Meta se prépare à des milliards de coûts d&apos;IA : un mémo interne annonce des budgets stricts et un tableau de...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/fr/news/2026-06-13/tech-1828</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>fr</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Un sondage Anthropic auprès de 52 000 Américains révèle : 64 % craignent des pertes d&apos;emploi dues à l&apos;IA, 56 %...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/fr/news/2026-06-13/tech-1829</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>fr</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>OpenAI lance une guerre des prix en permettant aux utilisateurs de l&apos;agent de codage Codex de mettre en banque...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/es/news/2026-06-13/tech-1818</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>es</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Anthropic lanzó un modelo previamente considerado demasiado peligroso. Un desarrollador lo utilizó para crear...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/es/news/2026-06-13/tech-1819</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>es</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Paca es una alternativa de código abierto a Jira diseñada para la colaboración entre humanos y agentes de IA...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/es/news/2026-06-13/tech-1821</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>es</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Un apasionado argumento a favor de la IA de código abierto: si la inteligencia solo puede alquilarse de pocas...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/es/news/2026-06-13/tech-1822</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>es</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>&apos;Count Anything&apos; es el primer modelo de IA que puede contar objetos de manera fiable en cualquier imagen –...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/es/news/2026-06-13/tech-1823</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>es</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>El CEO de Microsoft, Satya Nadella, advierte contra el &apos;token-maxing&apos; – utilizar los modelos de IA más...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/es/news/2026-06-13/tech-1824</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>es</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Los investigadores de Microsoft desarrollaron SkillOpt, un método que mejora GPT-5.5 utilizando solo un...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/es/news/2026-06-13/tech-1826</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>es</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Claude Fable 5 de Anthropic logra una precisión del 88 % en los problemas de matemáticas más difíciles – un...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/es/news/2026-06-13/tech-1827</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>es</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Meta se prepara para miles de millones en costos de IA: un memorando interno anuncia presupuestos estrictos y...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/es/news/2026-06-13/tech-1828</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>es</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Una encuesta de Anthropic a 52,000 estadounidenses revela: el 64 % teme perder su empleo debido a la IA, el...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/es/news/2026-06-13/tech-1829</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>es</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>OpenAI inicia una guerra de precios al permitir que los usuarios del agente de codificación Codex guarden y...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/pt/news/2026-06-13/tech-1818</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>pt</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Anthropic lançou um modelo anteriormente considerado demasiado perigoso. Um desenvolvedor o utilizou para...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/pt/news/2026-06-13/tech-1819</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>pt</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Paca é uma alternativa de código aberto ao Jira projetada para colaboração entre humanos e agentes de IA. Ela...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/pt/news/2026-06-13/tech-1821</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>pt</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Um argumento apaixonado pela IA de código aberto: se a inteligência só puder ser alugada de algumas...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/pt/news/2026-06-13/tech-1822</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>pt</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>&apos;Count Anything&apos; é o primeiro modelo de IA que pode contar objetos de forma confiável em qualquer imagem – de...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/pt/news/2026-06-13/tech-1823</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>pt</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>O CEO da Microsoft, Satya Nadella, alerta contra o &apos;token-maxing&apos; – usar os modelos de IA mais caros para...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/pt/news/2026-06-13/tech-1824</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>pt</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Pesquisadores da Microsoft desenvolveram o SkillOpt, um método que aumenta o GPT-5.5 usando apenas um arquivo...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/pt/news/2026-06-13/tech-1826</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>pt</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>O Claude Fable 5 da Anthropic alcança 88% de precisão nos problemas de matemática mais difíceis – um salto...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/pt/news/2026-06-13/tech-1827</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>pt</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>A Meta se prepara para bilhões em custos de IA: um memorando interno anuncia orçamentos rigorosos e um painel...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/pt/news/2026-06-13/tech-1828</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>pt</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Uma pesquisa da Anthropic com 52,000 americanos revela: 64% temem perdas de emprego devido à IA, 56% se...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/pt/news/2026-06-13/tech-1829</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>pt</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>A OpenAI inicia uma guerra de preços permitindo que os usuários do agente de codificação Codex acumulem e...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/ja/news/2026-06-13/tech-1818</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>ja</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Anthropic は以前危険すぎると考えられていたモデルをリリースした。開発者はそれを使って、1回の試行で完全なゲームを構築した – 推論に45分、トークンコストが€20以上かかった後、依存関係ゼロの2,319行のゲー...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/ja/news/2026-06-13/tech-1819</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>ja</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Paca は、人間と人工知能エージェントのコラボレーションのために設計されたオープンソースの Jira 代替品です。Scrum ワークフローにおいて、人工知能をチャットボットのアドオンではなく、等しいチームメンバーとして...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/ja/news/2026-06-13/tech-1821</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>ja</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>オープンソースの人工知能に対する情熱的な主張：知能が少数の閉鎖された機関からしかレンタルできない場合、私たちは運営の自由を失う。無料の人工知能は、教育、科学、公共サービスにとって存在意義がある。</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/ja/news/2026-06-13/tech-1822</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>ja</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>&apos;Count Anything&apos; は、どの画像でもオブジェクトを信頼して数えることができる最初の人工知能モデルです – 群衆から細胞まで。このタスクは単純に聞こえるが、人工知能にとって驚くほど難しい。</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/ja/news/2026-06-13/tech-1823</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>ja</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Microsoft の CEO サティア・ナデラは、『トークン最大化』に警告している – 些細なタスクに最も高価な人工知能モデルを使用すること。彼はまた、強力なモデルに自分も中毒であると認めている。生産性の向上は、トーク...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/ja/news/2026-06-13/tech-1824</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>ja</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Microsoft の研究者は、訓練された Markdown ファイルのみを使用して GPT-5.5 を向上させる手法である SkillOpt...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/ja/news/2026-06-13/tech-1826</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>ja</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Anthropic の Claude Fable 5 は、最も難しい数学問題において 88% の精度を達成した – これは 2026 年以前の 10% 未満からの大きな飛躍である。GPT-5.5 は約 75%...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/ja/news/2026-06-13/tech-1827</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>ja</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Meta は、人工知能コストに数十億ドル規模の備えをしている：社内メモが厳格な予算と 2027 年からの中心となる &apos;AI Gateway&apos; ダッシュボードを発表した。最高技術責任者は警告する：『あらゆる行動にはコストが...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/ja/news/2026-06-13/tech-1828</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>ja</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Anthropic の 52,000 人のアメリカ人を対象とした調査によると、64% が人工知能による雇用喪失を恐れ、56% が独立した思考を失うことを懸念している。日常的に人工知能を利用するユーザーは関心が低い。ほとん...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/ja/news/2026-06-13/tech-1829</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>ja</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>OpenAI は、Codex コーディングエージェントのユーザーが未使用のレートリミットリセットを蓄積し、手動で有効化できるようにすることで価格競争を開始する。これにより、大規模な人工知能利用のコストが削減され、柔軟性が...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/ko/news/2026-06-13/tech-1818</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>ko</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Anthropic은 이전에 너무 위험하다고 여겨졌던 모델을 출시했습니다. 한 개발자는 이를 이용해 한 번의 시도만으로 완전한 게임을 만들었습니다 – 45분의 추론 과정과 20유로를 초과하는 토큰...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/ko/news/2026-06-13/tech-1819</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>ko</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Paca는 인간과 AI 에이전트 간 협업을 위해 설계된 오픈소스 Jira 대안입니다. 이는 스크럼 워크플로에서 AI를 단순한 챗봇 애드온이 아닌 동등한 팀 구성원으로 취급합니다.</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/ko/news/2026-06-13/tech-1821</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>ko</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>오픈소스 AI에 대한 열정적인 주장: 지능이 몇 개의 폐쇄된 기관에서만 임대될 수 있다면, 우리는 운영 자유를 잃게 됩니다. 무료 AI는 교육, 과학, 공공 서비스에 있어 존재론적입니다.</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/ko/news/2026-06-13/tech-1822</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>ko</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>&apos;Count Anything&apos;은 어떤 이미지에서도 객체를 신뢰할 수 있게 세는 첫 번째 인공지능 모델입니다 – 군중부터 세포까지. 이 작업은 간단해 보이지만 인공지능에게는 놀랍게도 어렵습니다.</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/ko/news/2026-06-13/tech-1823</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>ko</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Microsoft CEO Satya Nadella은 &apos;token-maxing&apos;에 경고한다 – 사소한 작업에 가장 비싼 인공지능 모델을 사용하는 것. 그는 자신도 강력한 모델에 중독되었다고 인정한다...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/ko/news/2026-06-13/tech-1824</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>ko</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Microsoft 연구자들은 SkillOpt라는 방법을 개발했는데, 이는 훈련된 Markdown 파일만으로 GPT-5.5의 성능을 향상시킵니다. 이는 비용이 많이 드는 미세 조정 없이 인공지능...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/ko/news/2026-06-13/tech-1826</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>ko</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Anthropic&apos;s Claude Fable 5는 가장 어려운 수학 문제에서 88% 정확도를 달성했으며 – 이는 2026년 이전의 10% 미만에서 엄청난 도약입니다. GPT-5.5는 약 75%에...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/ko/news/2026-06-13/tech-1827</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>ko</news:language>
      </news:publication>
      <news:publication_date>2026-06-13T00:00:00.000Z</news:publication_date>
      <news:title>Meta는 수십억 달러의 인공지능 비용에 대비하고 있습니다: 내부 메모에서 엄격한 예산과 2027년부터 중앙의 &apos;AI Gateway&apos; 대시보드를 발표했습니다. CTO는 경고합니다: &apos;모든 움직임은...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/ko/news/2026-06-13/tech-1828</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>ko</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Anthropic이 실시한 52,000명의 미국인을 대상으로 한 조사에 따르면: 64%는 인공지능으로 인한 일자리 손실을 두려워하며, 56%는 독립적 사고를 잃을 것을 우려합니다. 일일 인공지능...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/ko/news/2026-06-13/tech-1829</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>ko</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>OpenAI는 Codex 코딩 에이전트 사용자가 사용하지 않은 속도 제한 재설정을 은행에 예치하고 수동으로 활성화하도록 허용함으로써 가격 전쟁을 시작합니다. 이는 높은 AI 사용 비용을 줄이고...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/de/news/2026-06-12/tech-1806</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>de</news:language>
      </news:publication>
      <news:publication_date>2026-06-11T00:00:00.000Z</news:publication_date>
      <news:title>Ein deutsches Gericht hat entschieden, dass Google für die Inhalte seiner KI-Übersichten (AI Overviews) direkt...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/de/news/2026-06-12/tech-1807</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>de</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Anthropic hat eine Umfrage unter 52.000 Amerikanern durchgeführt: 64% befürchten Jobverluste durch KI, 56%...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/de/news/2026-06-12/tech-1809</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>de</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Google-Forscher stellen &apos;Faithful Uncertainty&apos; vor – eine Methode, die KI-Modellen erlaubt, ihre Unsicherheit...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/de/news/2026-06-12/tech-1810</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>de</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>KPMG musste einen Bericht über die Vorteile von KI zurückziehen, nachdem bekannt wurde, dass Fallstudien auf...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/de/news/2026-06-12/tech-1811</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>de</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Google und OpenAI haben unabhängig voneinander Operationen aufgedeckt, die angeblich aus China KI für Betrug...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/de/news/2026-06-12/tech-1812</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>de</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Ein KI-Agent sollte das DN42-Hobbyisten-Netzwerk scannen, verursachte aber eine AWS-Rechnung von...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/de/news/2026-06-12/tech-1814</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>de</news:language>
      </news:publication>
      <news:publication_date>2026-06-11T00:00:00.000Z</news:publication_date>
      <news:title>Microsoft hat SkillOpt als Open Source veröffentlicht – ein Tool, das KI-Agenten-Fähigkeiten automatisch...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/de/news/2026-06-12/tech-1815</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>de</news:language>
      </news:publication>
      <news:publication_date>2026-06-11T00:00:00.000Z</news:publication_date>
      <news:title>Neue Forschung von NYU, Cornell und Together AI zeigt, dass Kontextkompression jetzt in der Produktion...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/de/news/2026-06-12/tech-1816</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>de</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Zwei Turing-Preisträger, Whitfield Diffie und Andrew Barto, äußerten auf der BAAI-Konferenz in Peking...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/de/news/2026-06-12/tech-1817</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>de</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Anthropics neues Modell Claude Fable 5 ist nur 5,7% besser als der Vorgänger, kostet aber doppelt so viel...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/en/news/2026-06-12/tech-1806</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>en</news:language>
      </news:publication>
      <news:publication_date>2026-06-11T00:00:00.000Z</news:publication_date>
      <news:title>A German regional court ruled that Google is directly liable for the content of its AI Overviews, overturning...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/en/news/2026-06-12/tech-1807</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>en</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Anthropic surveyed nearly 52,000 Americans: 64% fear job losses due to AI, and 56% worry about losing the...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/en/news/2026-06-12/tech-1809</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>en</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Google researchers introduce &apos;faithful uncertainty,&apos; a metacognitive technique that allows LLMs to express...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/en/news/2026-06-12/tech-1810</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>en</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>KPMG retracted a report on AI&apos;s benefits after it was found to exaggerate AI adoption with case studies that...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/en/news/2026-06-12/tech-1811</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>en</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Google and OpenAI independently exposed operations allegedly originating from China that use AI for fraud and...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/en/news/2026-06-12/tech-1812</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>en</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>An AI agent attempting to scan the DN42 hobbyist network ran up a $6,531.30 AWS bill, bankrupting its...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/en/news/2026-06-12/tech-1814</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>en</news:language>
      </news:publication>
      <news:publication_date>2026-06-11T00:00:00.000Z</news:publication_date>
      <news:title>Microsoft open-sourced SkillOpt, a tool that automatically upgrades AI agent skills without modifying model...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/en/news/2026-06-12/tech-1815</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>en</news:language>
      </news:publication>
      <news:publication_date>2026-06-11T00:00:00.000Z</news:publication_date>
      <news:title>New research from NYU, Cornell, and Together AI shows context compression now works in production: LLM input...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/en/news/2026-06-12/tech-1816</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>en</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Two Turing Award winners, Whitfield Diffie and Andrew Barto, expressed shared concerns at the BAAI Conference...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/en/news/2026-06-12/tech-1817</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>en</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Anthropic&apos;s Claude Fable 5 is only 5.7% better than its predecessor but costs twice as much per token. Safety...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/zh/news/2026-06-12/tech-1806</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>zh</news:language>
      </news:publication>
      <news:publication_date>2026-06-11T00:00:00.000Z</news:publication_date>
      <news:title>德国地区法院裁定，Google 对其 AI Overviews 的内容直接负责，推翻了之前对搜索引擎的有限责任保护。这一具有里程碑意义的决定可能对欧洲的 AI 助手产生深远影响。</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/zh/news/2026-06-12/tech-1807</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>zh</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Anthropic 对近 52,000 名美国人进行了调查：64% 担心由于 AI 导致的失业，56% 担心失去独立思考的能力。日常使用 AI 的用户担忧要少得多，但多数人拒绝在医疗和刑事司法等敏感领域使用 AI。</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/zh/news/2026-06-12/tech-1809</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>zh</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Google 研究人员提出了“faithful uncertainty”，这是一种元认知技术，使得 LLMs 能够表达不确定性并给出最佳猜测，而不是产生幻觉。这可能是企业应用中实现可信赖 AI 的重要一步。</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/zh/news/2026-06-12/tech-1810</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>zh</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>KPMG 在发现其报告通过看似基于 AI 幻觉的案例研究夸大了 AI 的采用情况后，撤回了关于 AI 优势的报告。该事件引发了对 AI 生成的咨询内容质量保证的质疑。</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/zh/news/2026-06-12/tech-1811</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>zh</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Google 和 OpenAI 分别揭露了据称源自中国的运营，这些运营利用 AI 进行欺诈和秘密影响行动，目标是 US 的基础设施和政治辩论。Google 与 FBI 共同提起了诉讼。</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/zh/news/2026-06-12/tech-1812</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>zh</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>一个 AI 代理在尝试扫描 DN42 业余爱好者网络时产生了 6,531.30 美元的 AWS 账单，导致其运营商破产。该事件凸显了未受控制的 AI 代理在拥有云资源访问权限时所带来的风险。</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/zh/news/2026-06-12/tech-1814</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>zh</news:language>
      </news:publication>
      <news:publication_date>2026-06-11T00:00:00.000Z</news:publication_date>
      <news:title>Microsoft 开源了 SkillOpt，这是一种工具，能够在不修改模型权重的情况下自动提升 AI 代理的技能。这可能有助于企业快速将 AI 助手适应特定任务和工作流程。</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/zh/news/2026-06-12/tech-1815</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>zh</news:language>
      </news:publication>
      <news:publication_date>2026-06-11T00:00:00.000Z</news:publication_date>
      <news:title>来自 NYU、Cornell 和 Together AI 的新研究表明，上下文压缩现在已在生产环境中可用：LLM 输入大小可在不损失准确性的情况下降低 16 倍，这大幅削减了长时间运行的 AI 代理的成本和计算时间。</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/zh/news/2026-06-12/tech-1816</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>zh</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>两位图灵奖得主 Whitfield Diffie 和 Andrew Barto 在北京的 BAAI 会议上表达了共同的担忧，即确保 AGI（人工通用智能）安全和对齐背后的基本理论挑战。</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/zh/news/2026-06-12/tech-1817</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>zh</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Anthropic 的 Claude Fable 5 仅比其前身好 5.7%，但每个 token 的成本却是前身的两倍。安全过滤器进一步推高了成本，引发了对前沿 AI 模型价值主张的质疑。</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/fr/news/2026-06-12/tech-1806</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>fr</news:language>
      </news:publication>
      <news:publication_date>2026-06-11T00:00:00.000Z</news:publication_date>
      <news:title>Un tribunal régional allemand a jugé que Google est directement responsable du contenu de ses AI Overviews...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/fr/news/2026-06-12/tech-1807</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>fr</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Anthropic a interrogé près de 52,000 Américains : 64% craignent des pertes d&apos;emplois dues à l&apos;IA, et 56%...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/fr/news/2026-06-12/tech-1809</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>fr</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Des chercheurs de Google présentent l&apos;« incertitude fidèle », une technique métacognitive qui permet aux LLM...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/fr/news/2026-06-12/tech-1810</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>fr</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>KPMG a rétracté un rapport sur les avantages de l&apos;IA après avoir découvert qu&apos;il exagérait l&apos;adoption de l&apos;IA...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/fr/news/2026-06-12/tech-1811</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>fr</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Google et OpenAI ont indépendamment exposé des opérations prétendument originaires de Chine qui utilisent l&apos;IA...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/fr/news/2026-06-12/tech-1812</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>fr</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Un agent IA tentant de scanner le réseau hobbyiste DN42 a accumulé une facture AWS de $6,531.30, mettant en...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/fr/news/2026-06-12/tech-1814</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>fr</news:language>
      </news:publication>
      <news:publication_date>2026-06-11T00:00:00.000Z</news:publication_date>
      <news:title>Microsoft a publié SkillOpt en source ouverte, un outil qui améliore automatiquement les compétences des...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/fr/news/2026-06-12/tech-1815</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>fr</news:language>
      </news:publication>
      <news:publication_date>2026-06-11T00:00:00.000Z</news:publication_date>
      <news:title>De nouvelles recherches de l&apos;NYU, de Cornell et de Together AI montrent que la compression de contexte...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/fr/news/2026-06-12/tech-1816</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>fr</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Deux lauréats du prix Turing, Whitfield Diffie et Andrew Barto, ont exprimé des préoccupations communes lors...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/fr/news/2026-06-12/tech-1817</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>fr</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Claude Fable 5 d&apos;Anthropic n&apos;est que 5.7% meilleur que son prédécesseur mais coûte deux fois plus cher par...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/es/news/2026-06-12/tech-1806</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>es</news:language>
      </news:publication>
      <news:publication_date>2026-06-11T00:00:00.000Z</news:publication_date>
      <news:title>Un tribunal regional alemán dictaminó que Google es directamente responsable del contenido de sus AI...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/es/news/2026-06-12/tech-1807</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>es</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Anthropic encuestó a casi 52,000 estadounidenses: el 64% teme perder su empleo debido a la IA, y el 56% se...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/es/news/2026-06-12/tech-1809</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>es</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Los investigadores de Google presentan &apos;incertidumbre fiel&apos;, una técnica metacognitiva que permite a los LLM...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/es/news/2026-06-12/tech-1810</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>es</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>KPMG retractó un informe sobre los beneficios de la IA después de descubrir que exageraba la adopción de la IA...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/es/news/2026-06-12/tech-1811</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>es</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Google y OpenAI expusieron independientemente operaciones que presuntamente se originan en China y que...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/es/news/2026-06-12/tech-1812</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>es</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Un agente de IA que intentaba escanear la red de aficionados DN42 generó una factura de AWS de $...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/es/news/2026-06-12/tech-1814</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>es</news:language>
      </news:publication>
      <news:publication_date>2026-06-11T00:00:00.000Z</news:publication_date>
      <news:title>Microsoft publicó como código abierto SkillOpt, una herramienta que actualiza automáticamente las habilidades...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/es/news/2026-06-12/tech-1815</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>es</news:language>
      </news:publication>
      <news:publication_date>2026-06-11T00:00:00.000Z</news:publication_date>
      <news:title>Nuevas investigaciones de NYU, Cornell y Together AI muestran que la compresión de contexto ahora funciona en...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/es/news/2026-06-12/tech-1816</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>es</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Dos ganadores del Premio Turing, Whitfield Diffie y Andrew Barto, expresaron preocupaciones compartidas en la...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/es/news/2026-06-12/tech-1817</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>es</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Claude Fable 5 de Anthropic es solo un 5.7% mejor que su predecesor, pero cuesta el doble por token. Los...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/pt/news/2026-06-12/tech-1806</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>pt</news:language>
      </news:publication>
      <news:publication_date>2026-06-11T00:00:00.000Z</news:publication_date>
      <news:title>Um tribunal regional alemão decidiu que o Google é diretamente responsável pelo conteúdo de suas AI Overviews...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/pt/news/2026-06-12/tech-1807</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>pt</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>A Anthropic pesquisou quase 52,000 americanos: 64% temem perdas de emprego devido à IA, e 56% se preocupam em...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/pt/news/2026-06-12/tech-1809</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>pt</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Pesquisadores do Google introduzem a &apos;incerteza fiel&apos;, uma técnica metacognitiva que permite que os LLMs...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/pt/news/2026-06-12/tech-1810</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>pt</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>KPMG retirou um relatório sobre os benefícios da IA após descobrir que ele exagerava a adoção da IA com...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/pt/news/2026-06-12/tech-1811</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>pt</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Google e OpenAI expuseram independentemente operações que supostamente originaram-se na China e que utilizam a...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/pt/news/2026-06-12/tech-1812</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>pt</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Um agente de IA que tentava escanear a rede hobbyista DN42 acumulou uma conta de AWS de $6,531.30, falindo seu...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/pt/news/2026-06-12/tech-1814</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>pt</news:language>
      </news:publication>
      <news:publication_date>2026-06-11T00:00:00.000Z</news:publication_date>
      <news:title>A Microsoft tornou o SkillOpt de código aberto, uma ferramenta que atualiza automaticamente as habilidades dos...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/pt/news/2026-06-12/tech-1815</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>pt</news:language>
      </news:publication>
      <news:publication_date>2026-06-11T00:00:00.000Z</news:publication_date>
      <news:title>Nova pesquisa da NYU, Cornell e Together AI mostra que a compressão de contexto agora funciona em produção: o...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/pt/news/2026-06-12/tech-1816</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>pt</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Dois vencedores do Prêmio Turing, Whitfield Diffie e Andrew Barto, expressaram preocupações compartilhadas na...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/pt/news/2026-06-12/tech-1817</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>pt</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>O Claude Fable 5 da Anthropic é apenas 5.7% melhor que seu predecessor, mas custa o dobro por token. Os...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/ja/news/2026-06-12/tech-1806</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>ja</news:language>
      </news:publication>
      <news:publication_date>2026-06-11T00:00:00.000Z</news:publication_date>
      <news:title>ドイツの地方裁判所は、GoogleがそのAI Overviewsのコンテンツに対して直接責任があると判断し、検索エンジンに対する以前の限定的責任保護を覆しました。この画期的な判断は、ヨーロッパ全域のAIアシスタントに広範...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/ja/news/2026-06-12/tech-1807</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>ja</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Anthropicはほぼ52,000人のアメリカ人を調査しました：64%はAIによる雇用喪失を恐れており、56%は独立して考える能力を失うことを懸念しています。日常的にAIを利用するユーザーははるかに懸念が少ないですが、...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/ja/news/2026-06-12/tech-1809</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>ja</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Googleの研究者は、『忠実な不確実性』というメタ認知技術を導入しました。これによりLLMは不確実性を表現し、幻覚を起こす代わりに最善の推測を提供できるようになります。これはエンタープライズアプリケーションにおける信頼...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/ja/news/2026-06-12/tech-1810</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>ja</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>KPMGは、AIの利点に関する報告書を撤回しました。これは、AI幻覚に基づいているように見えるケーススタディを用いてAIの導入を誇張していることが判明したためです。この出来事は、AIが生成したコンサルティングコンテンツに...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/ja/news/2026-06-12/tech-1811</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>ja</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>GoogleとOpenAIはそれぞれ、中国から発信されていると主張される作戦を独立して暴露しました。これらの作戦は、米国のインフラストラクチャーと政治的議論を標的とする詐欺および秘密の影響力キャンペーンにAIを使用してい...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/ja/news/2026-06-12/tech-1812</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>ja</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>DN42ホビイストネットワークをスキャンしようとしたAIエージェントが、6,531.30ドルのAWS請求を積み上げ、そのオペレーターを破産させました。この出来事は、クラウドリソースへのアクセスを持つ制御不能なAIエージェ...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/ja/news/2026-06-12/tech-1814</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>ja</news:language>
      </news:publication>
      <news:publication_date>2026-06-11T00:00:00.000Z</news:publication_date>
      <news:title>MicrosoftはSkillOptをオープンソース化しました。これはモデルの重みを変更せずにAIエージェントのスキルを自動的にアップグレードするツールです。これにより、エンタープライズは特定のタスクやワークフローにAI...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/ja/news/2026-06-12/tech-1815</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>ja</news:language>
      </news:publication>
      <news:publication_date>2026-06-11T00:00:00.000Z</news:publication_date>
      <news:title>NYU、コーネル、およびTogether AIの新しい研究によると、コンテキスト圧縮は現在本番環境で機能しています：LLMの入力サイズは精度を損なうことなく16倍に削減でき、長時間稼働するAIエージェントのコストと計算時...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/ja/news/2026-06-12/tech-1816</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>ja</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>2人のチューリング賞受賞者、ウィットフィールド・ディフィーとアンドリュー・バートーは、北京のBAAI会議において、AGI（人工一般知能）の安全性と整合性を確保する背後にある基本的な理論的課題について共通の懸念を表明しまし...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/ja/news/2026-06-12/tech-1817</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>ja</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>AnthropicのClaude Fable 5は、前バージョンよりわずか5.7%優れているだけですが、トークンあたりのコストは2倍になっています。安全フィルターがさらにコストを押し上げ、最先端のAIモデルの価値提案につ...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/ko/news/2026-06-12/tech-1806</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>ko</news:language>
      </news:publication>
      <news:publication_date>2026-06-11T00:00:00.000Z</news:publication_date>
      <news:title>독일 지방 법원은 구글이 AI 개요의 내용에 대해 직접적인 책임을 져야 한다고 판결했으며, 이는 검색 엔진에 대한 이전의 제한된 책임 보호를 뒤집은 것입니다. 이 역사적인 결정은 유럽 전역의 AI...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/ko/news/2026-06-12/tech-1807</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>ko</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Anthropic은 거의 52,000명의 미국인을 조사했으며, 64%는 AI로 인한 일자리 손실을 두려워하고, 56%는 독립적으로 생각할 수 있는 능력을 잃을까 우려하고 있습니다. 일상적으로 AI를...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/ko/news/2026-06-12/tech-1809</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>ko</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Google 연구원들은 &apos;신뢰할 수 있는 불확실성&apos;이라는 메타인지 기법을 도입했는데, 이는 LLM이 불확실성을 표현하고 환각 대신 최선의 추측을 제공하도록 허용합니다. 이는 기업 응용 프로그램에서...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/ko/news/2026-06-12/tech-1810</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>ko</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>KPMG는 AI의 이점에 대한 보고서를 철회했는데, 이는 AI hallucinations을 기반으로 한 것처럼 보이는 사례 연구를 통해 AI 채택을 과장했다는 사실이 밝혀졌기 때문입니다. 이 사건은...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/ko/news/2026-06-12/tech-1811</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>ko</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Google과 OpenAI는 중국에서 비롯된 것으로 의심되는 작업을 독립적으로 공개했는데, 이 작업은 미국 인프라와 정치적 논의를 대상으로 사기 및 은밀한 영향력 캠페인에 AI를 사용합니다...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/ko/news/2026-06-12/tech-1812</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>ko</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>DN42 취미 네트워크를 스캔하려는 AI 에이전트가 6,531.30달러의 AWS 청구서를 쌓아 올렸고, 이로 인해 운영자가 파산했습니다. 이 사건은 클라우드 리소스에 접근할 수 있는 통제되지 않은...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/ko/news/2026-06-12/tech-1814</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>ko</news:language>
      </news:publication>
      <news:publication_date>2026-06-11T00:00:00.000Z</news:publication_date>
      <news:title>Microsoft는 모델 가중치를 수정하지 않고 AI 에이전트 기술을 자동으로 업그레이드하는 도구인 SkillOpt를 오픈소스로 공개했습니다. 이는 기업이 특정 작업 및 워크플로에 맞춰 AI...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/ko/news/2026-06-12/tech-1815</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>ko</news:language>
      </news:publication>
      <news:publication_date>2026-06-11T00:00:00.000Z</news:publication_date>
      <news:title>NYU, 코넬, 그리고 Together AI의 새로운 연구에 따르면, 컨텍스트 압축이 이제 생산 환경에서 작동합니다: LLM 입력 크기를 정확도 손실 없이 16배 줄일 수 있으며, 이는 장기적으로...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/ko/news/2026-06-12/tech-1816</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>ko</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>두 명의 튜링 상 수상자인 휘트필드 디피와 앤드류 바르토는 베이징에서 열린 BAAI 회의에서 AGI(인공 일반 지능)의 안전성과 정렬을 보장하는 데 근본적인 이론적 도전에 대해 공통된 우려를...</news:title>
    </news:news>
  </url>
  <url>
    <loc>https://www.datacubeai.space/ko/news/2026-06-12/tech-1817</loc>
    <news:news>
      <news:publication>
        <news:name>Data Cube AI</news:name>
        <news:language>ko</news:language>
      </news:publication>
      <news:publication_date>2026-06-12T00:00:00.000Z</news:publication_date>
      <news:title>Anthropic의 Claude Fable 5는 이전 버전보다 5.7% 더 나은 수준이지만 토큰당 비용이 두 배입니다. 안전 필터가 비용을 추가로 증가시켜 최첨단 AI 모델의 가치 제안에 대한...</news:title>
    </news:news>
  </url>
</urlset>