![[f57c41fa-a786-4416-8459-fdca8ec30f13.mp3]]

库克动员员工：苹果将以自研芯片为核心把握AI机遇

简报：

• 苹果公司首席执行官库克在全体员工大会上表示，公司必须抓住并把握人工智能（AI）这一机遇。
• 库克强调，苹果自研芯片将是其AI战略的关键，公司将为此投入资源以实现目标。

相关链接：

• https://www.36kr.com/newsflashes/3405167133363588

Cook Rallies Employees: Apple to Seize AI Opportunity with In-house Chips at Its Core 🚀

Brief:

• Apple CEO Tim Cook stated at an all-hands meeting that the company must grasp and embrace the Artificial Intelligence (AI) opportunity. 💡
• Cook emphasized that Apple's self-developed chips will be crucial to its AI strategy, and the company will invest resources to achieve this goal. 🧠

Related Links:

• https://www.36kr.com/newsflashes/3405167133363588

Poe平台开放API，订阅即可调用全部AI模型

简报：

• AI聊天平台Poe近日宣布推出API功能，平台订阅用户无需额外付费即可直接调用。
• 该API支持访问Poe平台上的所有AI模型和机器人，范围覆盖文本、图像及视频模型。
• 为方便开发者集成，Poe的API兼容OpenAI的chat completions接口，降低了技术迁移成本。

相关链接：

• Poe震撼升级：API开放，订阅即用，图像视频AI模型全覆盖，兼容OpenAI接口！

Poe Platform Unveils API: All AI Models Accessible with Subscription 🎉

Brief:

• AI chat platform Poe recently announced the launch of an API feature, enabling platform subscribers to access it directly without additional fees.
• This API provides access to all AI models and bots available on the Poe platform, encompassing text, image, and video models. 🚀
• For seamless developer integration, Poe's API is compatible with OpenAI's chat completions interface, thus lowering technical migration costs. 🤝

Related Link:

• Poe's Tremendous Upgrade: API Now Open, Subscription-Ready, Full Coverage of Image & Video AI Models, Compatible with OpenAI Interface!

OpenAI视AI承认解题失败为重要突破

简报：

• OpenAI的AI模型在挑战国际数学奥林匹克竞赛（IMO）第六题时，因无法解答而坦承“我无法回答”，最终得分零分。
• OpenAI研究员Noam Brown和Alex Wei认为，模型在缺乏有效证据时承认“不知道”，是其具备“自我意识”的体现。
• 这一转变被视为重要的技术突破，标志着模型开始从生成看似可信但错误的“幻觉”答案，转向更加可靠、能认识到自身局限性的新阶段。
• 尽管在此难题上失败，但由三名研究员组成的团队仅用两个月就达成了为AI摘得IMO金牌的目标。

相关链接：

• AI 答 IMO 难题坦承“不会”，OpenAI 称“这就是自我意识”

OpenAI Views AI Admitting Failure to Solve Problems as a Significant Breakthrough ✨

Briefing:

• OpenAI's AI model openly admitted "I cannot answer" when challenged with the sixth problem of the International Mathematical Olympiad (IMO) due to its inability to solve it, ultimately scoring zero points.
• OpenAI researchers Noam Brown and Alex Wei believe that the model's admission of "I don't know" when lacking sufficient evidence is an embodiment of its "self-awareness." 🧠
• This shift is regarded as a significant technological breakthrough, marking a new stage where the model begins to move away from generating seemingly plausible but incorrect "hallucinated" answers towards a more reliable state, recognizing its own limitations.
• Although it failed on this particular difficult problem, a team of three researchers accomplished the goal of winning an IMO gold medal for AI in just two months. 👍

Related Links:

• AI Answers IMO Difficult Problem, Admits "I Don't Know," OpenAI Calls This "Self-Awareness"

OpenAI因违规使用竞品，被Anthropic撤销Claude API权限

简报：

• 在新模型GPT-5发布前夕，Anthropic公司指控OpenAI违反其服务条款，并已撤销OpenAI对其模型Claude的API访问权限。
• Anthropic称，OpenAI使用其服务来“构建竞争性产品或服务”，这直接违反了其商业服务条款。
• OpenAI回应称，评估其他AI系统以对比进度和提高安全性是行业惯例，并对Anthropic的决定表示失望，指出对方仍可使用OpenAI的API。
• 据悉，OpenAI通过API接入Claude，以评估其在编程、创意写作及安全响应等方面的能力。

相关链接：

• GPT-5 发布在即：OpenAI 被控违反服务条款，Claude API 访问权限被撤销
• AI 答 IMO 难题坦承“不会”，OpenAI 称“这就是自我意识”
• OpenAI 首个开源模型？神秘 Horizon Alpha 崭露头角，超 Kimi K2 登顶 EQ-Bench 创意写作排行榜
• 全球估值最高 AI 公司：消息称 OpenAI 最新融资超 80 亿美元，提前完成今年 400 亿美元目标
• 多项力压 Grok 4、OpenAI o3，谷歌推出 Gemini 2.5 Deep Think 模型

OpenAI's Claude API Access Revoked by Anthropic Over Alleged Terms of Service Violation 🚫

Brief:

• On the eve of its new model GPT-5's release, Anthropic has accused OpenAI of violating its service terms and has consequently revoked OpenAI's API access to its Claude model.
• Anthropic claims OpenAI used its service to "build competitive products or services," a direct violation of its commercial terms of service. 🤖
• OpenAI responded by stating that evaluating other AI systems to compare progress and enhance safety is standard industry practice, expressing disappointment with Anthropic's decision, and noting that Anthropic can still use OpenAI's APIs. 🤔
• It is understood that OpenAI accessed Claude via its API to evaluate its capabilities in areas such as programming, creative writing, and safety responses.

Related Links:

• GPT-5 Release Imminent: OpenAI Accused of Violating Terms of Service, Claude API Access Revoked
• AI Answers IMO Problem, Admits "Cannot Solve It," OpenAI States "This Is Self-Awareness"
• OpenAI's First Open-Source Model? Mysterious Horizon Alpha Emerges, Surpasses Kimi K2 to Top EQ-Bench Creative Writing Rankings
• World's Highest-Valued AI Company: OpenAI's Latest Funding Exceeds $8 Billion, Reportedly Achieving This Year's$40 Billion Target Ahead of Schedule
• Outperforming Grok 4 and OpenAI o3 in Multiple Aspects, Google Launches Gemini 2.5 Deep Think Model

昆仑万维开源推理大模型MindLink，主打自适应推理与降本增效

简报：

• 昆仑万维于8月2日发布并开源了全新的推理大模型Skywork MindLink。
• 该模型基于Qwen3-32B和Qwen2.5-72B进行后训练，旨在节省训练成本并提升模型能力。
• 核心创新是引入“Plan-based Reasoning”推理范式，能根据任务难度自适应整合推理与非推理回复，以降低成本、提升效率。
• 在多个大模型评测中表现优异，尤其在奥赛等复杂数学领域以及GPQA-diamond等通识类评测中获得多项第一。
• 72B模型权重、技术报告及代码仓库均已开源。

相关链接：

• 模型权重
• 技术报告
• 代码仓库

Kunlun Tech Open-Sources Inference LLM MindLink, Focusing on Adaptive Reasoning and Cost-Efficiency 🚀

Brief:

• Kunlun Tech officially released and open-sourced its new inference large language model, Skywork MindLink, on August 2nd.
• The model is fine-tuned based on Qwen3-32B and Qwen2.5-72B, aiming to significantly reduce training costs while enhancing model capabilities.
• A core innovation is the introduction of the "Plan-based Reasoning" inference paradigm ✨, which adaptively integrates reasoning and non-reasoning responses based on task difficulty to lower costs and improve efficiency.
• MindLink has demonstrated outstanding performance in multiple large model evaluations, securing several first-place rankings, especially in complex mathematical domains like Olympiads and general knowledge benchmarks such as GPQA-diamond.
• The 72B model weights, technical report, and code repository have all been open-sourced. 🤝

Related Links:

• Model Weights
• Technical Report
• Code Repository

谷歌发布 Gemini 2.5 Deep Think，基准测试超越 OpenAI o3

简报：

• 谷歌于8月1日宣布，Gemini 2.5 Deep Think 模型已正式上线，并向 Gemini 应用的 Google AI Ultra 订阅用户开放。
• 该模型采用全新的“Deep Think”推理模式，可在作答前综合多个假设进行推理，并自动集成代码执行和谷歌搜索等工具。
• 在LiveCodeBench V6、Humanity’s Last Exam等多项关键基准测试中，Gemini 2.5 Deep Think 的表现已超越 OpenAI 的 o3 和 xAI 的 Grok 4。
• 谷歌计划在未来几周内，通过 Gemini API 向部分开发者提供该模型的不同版本。

相关链接：

• 多项力压 Grok 4、OpenAI o3，谷歌推出 Gemini 2.5 Deep Think 模型
• 谷歌推出Gemini 2.5 Deep Think模型，多项力压Grok 4、OpenAI o3

Google Releases Gemini 2.5 Deep Think, Outperforming OpenAI o3 in Benchmarks

Brief:

• On August 1st, Google announced that the Gemini 2.5 Deep Think model has officially launched and is now available to Google AI Ultra subscribers of the Gemini application.
• The model features a new “Deep Think” 🤔 reasoning mode, allowing it to synthesize multiple hypotheses for reasoning before responding, and automatically integrating tools like code execution and Google Search.
• In multiple key benchmarks, including LiveCodeBench V6 and Humanity’s Last Exam, Gemini 2.5 Deep Think has outperformed OpenAI's o3 and xAI's Grok 4. 🚀
• Google plans to offer different versions of the model to select developers via the Gemini API in the coming weeks. ✨

Related Links:

• Google Launches Gemini 2.5 Deep Think Model, Outperforming Grok 4 and OpenAI o3 in Multiple Aspects
• Google Launches Gemini 2.5 Deep Think Model, Outperforming Grok 4 and OpenAI o3 in Multiple Benchmarks

Creao AI 获数千万美元融资，开发 Agent 协同操作系统

简报：

• AI初创公司 Creao AI 宣布于2025年7月31日连续完成两轮融资，包括一轮数百万美元的天使轮和一轮数千万美元的Pre-A轮。
• 本轮融资由Yunqi、Monolith、GL Ventures、HongShan、Hua Capital等顶级风投机构参与。
• 资金将用于构建其核心产品——Agentic OS（AOS），一个以AI Agent为核心的下一代操作系统。
• AOS旨在通过简单的对话即可生成并部署完整的全栈Agentic App，打通独立Agent的协同与调度，构建高效的数字生产网络。

相关链接：

• https://www.aibase.com/zh/news/20168

Creao AI Secures Multi-Million Dollar Funding for Agent Collaborative Operating System Development

Briefing:

• AI startup Creao AI announced on July 31, 2025, the consecutive completion of two funding rounds: a multi-million dollar Angel Round and a multi-ten-million dollar Pre-A Round.
• This funding round saw participation from top-tier venture capital firms including Yunqi, Monolith, GL Ventures, HongShan, and Hua Capital. 🚀
• The capital will be allocated to building its core product – Agentic OS (AOS) – a next-generation operating system centered around AI Agents. ✨
• AOS aims to generate and deploy complete full-stack Agentic Apps through simple conversations, enabling collaboration and scheduling among independent Agents, thereby building an efficient digital production network. 💡

Related Link:

• https://www.aibase.com/zh/news/20168