![[b268e3f5-a2c4-40bc-86a8-3c0edf739948.mp3]]

阿里Qwen3-Coder性能超越竞品，将赋能惠普智能助手

简报：

• 阿里巴巴的Qwen3-Coder在关键基准测试中，表现优于DeepSeek和月之暗面（Moonshot）的K2等中国竞争对手。
• 该模型即将为惠普的“小微慧”（Xiaowei Hui）智能助手提供技术支持。

Alibaba Qwen3-Coder Outperforms Competitors, Set to Power HP Smart Assistant

Briefing:

• Alibaba's Qwen3-Coder has outperformed Chinese competitors like DeepSeek and Moonshot's K2 in key benchmark tests. 🚀
• The model is poised to power HP's 'Xiaowei Hui' smart assistant. 🤖✨

GPT-5用户反馈不佳，OpenAI重启旧版模型以应对

简报：

• OpenAI的ChatGPT负责人Nick Turley表示，因新一代模型GPT-5的推出过程不顺利，公司决定重新开放用户对旧版GPT-4o的访问权限。
• Turley称，用户对GPT-4o的喜爱和依赖程度让他感到意外，而GPT-5“较少阿谀奉承”的特性未得到用户广泛认可，使用者感受到了较大的体验落差。
• OpenAI强调用户反馈的重要性，未来将更注重与用户的沟通，以平衡技术改进与用户体验。

相关链接：

• OpenAI 高管对用户对 GPT-4o 的热情感到意外，重启版本以应对反馈

GPT-5 Faces Poor User Feedback, OpenAI Reinstates Older Model Access

Brief:

• Nick Turley, Head of ChatGPT at OpenAI, stated that due to a bumpy rollout of the new-generation GPT-5 model, the company has decided to re-enable user access to the older GPT-4o version. 🔄
• Turley mentioned that he was surprised by the level of user affection and reliance on GPT-4o, and that GPT-5's "less sycophantic" nature was not widely accepted by users, who felt a significant drop in experience. 😮
• OpenAI emphasized the importance of user feedback, stating that in the future, it will focus more on communication with users to balance technological advancements with user experience. 🤝

Related Links:

• OpenAI Executive Surprised by User Enthusiasm for GPT-4o, Reverts Version to Address Feedback

OpenAI高管：GPT-5仍会编造信息，用户需自行核实

简报：

• OpenAI ChatGPT负责人尼克・特利（Nick Turley）在接受采访时承认，新模型GPT-5仍然存在“幻觉”（编造信息）问题，在可靠性上远未达到完美。
• 他建议用户将ChatGPT的回答作为“第二种参考意见”而非主要事实来源，并敦促用户务必仔细核对模型给出的答案。
• 特利表示，尽管GPT-5在减少“幻觉”方面取得了“巨大进步”，但他预计该问题不会在未来一个季度内被完全解决。
• 他认为，将大型语言模型与搜索引擎等事实来源相结合是目前最佳的产品形态，这也是ChatGPT引入搜索功能的原因。

相关链接：

• ChatGPT 负责人坦言：GPT-5 仍有“幻觉”问题，建议用户核对答案

OpenAI Executive: GPT-5 Will Still Fabricate Information, Users Must Verify ⚠️

Briefing:

• Nick Turley, Head of ChatGPT at OpenAI, admitted in an interview that the new model GPT-5 still suffers from "hallucinations" (fabricating information) and is far from perfect in terms of reliability.
• He advised users to treat ChatGPT's answers as a "second opinion" rather than a primary source of facts, and urged them to carefully verify the information provided by the model. 🧐
• Turley stated that despite GPT-5 making "significant progress" in reducing "hallucinations," he does not expect the issue to be fully resolved within the next quarter.
• He believes that combining large language models with factual sources like search engines represents the best current product form, which is why ChatGPT introduced search functionality. 💡

Related Link:

• ChatGPT Head Admits: GPT-5 Still Has "Hallucination" Issues, Advises Users to Verify Answers

奥尔特曼复盘GPT-5发布失误，并称未来CEO或由AI担任

简报：

• OpenAI CEO萨姆·奥尔特曼在接受采访时承认，公司在近期GPT-5的发布中“确实搞砸了一些事情”，主要指新模型移除了用户习惯的GPT-4o的“温暖”特质，引发部分用户抗议，公司已迅速将4o作为可选模型重新上线。
• 尽管发布存在争议，奥尔特曼透露公司API流量在48小时内翻倍，GPU已不够用，ChatGPT日活用户数持续创新高。
• 当被问及未来是否会继续担任CEO时，奥尔特曼表示，或许三年后这个职位将由AI接手。
• 奥尔特曼还谈及了当前行业存在AI泡沫、公司面临GPU算力严重不足以及计划资助脑机接口公司等多个话题。
• 另外，ChatGPT负责人尼克·特利也坦言，GPT-5模型仍存在“幻觉”问题，建议用户在使用时仔细核对答案。

相关链接：

• OpenAI 奥尔特曼：GPT-5 有点搞砸了，未来公司 CEO 或是 AI
• ChatGPT 负责人坦言：GPT-5 仍有“幻觉”问题，建议用户核对答案

Altman Reflects on GPT-5 Launch Mishaps, Suggests Future CEOs Could Be AI

Briefing:

• OpenAI CEO Sam Altman admitted in an interview that the company "really messed some things up" 🤔 with the recent GPT-5 launch, primarily referring to the new model's removal of the "warm" characteristic users were accustomed to from GPT-4o, which sparked protests from some users. The company has since quickly re-launched 4o as an optional model.
• Despite the controversial launch, Altman revealed that the company's API traffic doubled within 48 hours, GPUs are no longer sufficient, and ChatGPT's daily active user count continues to hit new highs 🚀.
• When asked if he would continue to serve as CEO in the future, Altman stated that perhaps in three years, this position might be taken over by AI 🤖.
• Altman also discussed various topics, including the current AI bubble in the industry, the company's severe shortage of GPU computing power, and plans to fund brain-computer interface companies.
• Additionally, Nick T V, Head of ChatGPT, admitted that the GPT-5 model still suffers from "hallucination" issues and advised users to carefully verify answers when using it.

Related Links:

• OpenAI's Sam Altman: GPT-5 Was a Bit Messed Up, Future CEO Might Be AI
• ChatGPT Head Admits: GPT-5 Still Has 'Hallucination' Issues, Advises Users to Verify Answers

宇树G1人形机器人在世界机器人运动会百米障碍赛夺冠

在首届世界人形机器人运动会上，宇树科技的G1人形机器人在100米障碍赛中以33.71秒的成绩夺得金牌。

相关链接：

• 宇树科技 100 米障碍赛再摘金牌，G1 人形机器人夺冠成绩 33.71 秒
• 0.95 米，星动纪元星动 L7 获首届世界人形机器人运动会跳高冠军
• 穿越 2500 年的赛博礼乐：人形机器人首次用编钟弹奏老祖宗的 BGM

Unitree G1 Humanoid Robot Wins 100-Meter Obstacle Course at World Robot Games 🥇

At the inaugural World Humanoid Robot Games, Unitree Robotics' G1 humanoid robot clinched the gold medal in the 100-meter obstacle course with a remarkable time of 33.71 seconds. What an achievement! 🏆🤖

Related Links:

• Unitree Robotics Claims Another Gold Medal in 100-Meter Obstacle Course, G1 Humanoid Robot Wins with 33.71 Seconds
• 0.95 Meters, Starry Robotics Starry L7 Wins High Jump Championship at Inaugural World Humanoid Robot Games
• Cyber Rites and Music Spanning 2500 Years: Humanoid Robots Play Ancestral BGM on Chime Bells for the First Time

数据量激增，IDC预测中国AI硬件支出2028年将达330亿元

简报：

• 根据国际数据公司（IDC）预测，受大模型商用落地驱动，中国生成式AI相关硬件支出将从2023年的65亿元增长至2028年的330亿元，年复合增长率达38.5%。
• 报告指出，数据消耗量呈爆发式增长，预计到2024年下半年，中国日均Token消耗量将达到114.2万亿。
• 为应对数据流量压力，高端网络硬件市场迎来快速发展，预计到2029年中国高端以太网（≥200G）端口出货量将超过4300万个。

相关链接：

• IDC：未来五年，中国生成式 AI 相关硬件支出将激增至 330 亿元
• 阿里启动近千人AI招聘计划 重点布局大模型与AI硬件
• Arm 推出 AI 驱动技术，提升移动设备图形质量
• 台灣製造全球首發！HTC推AI眼鏡VIVE Eagle 功能、價格、上市日一次看
• AI眼鏡複製iPhone爆發路徑？科技巨頭搶攻穿戴新藍海

Data Volume Surges, IDC Forecasts China's AI Hardware Spending to Reach 33 Billion Yuan by 2028 🚀

Brief:

• According to International Data Corporation (IDC) forecasts, driven by the commercial deployment of large models, China's spending on generative AI-related hardware is expected to grow from 6.5 billion yuan in 2023 to 33 billion yuan by 2028, representing a Compound Annual Growth Rate (CAGR) of 38.5%.
• The report highlights an explosive growth in data consumption, predicting that by the second half of 2024, China's average daily token consumption will reach 114.2 trillion.
• To cope with the data traffic pressure, the high-end networking hardware market is experiencing rapid development. It is projected that by 2029, China's shipments of high-end Ethernet ports (≥200G) will exceed 43 million units. 🌐

Related Links: 💡

• IDC: China's Generative AI Hardware Spending to Soar to 33 Billion Yuan in Next Five Years
• Alibaba Launches Nearly Thousand-Person AI Recruitment Plan, Focusing on Large Models and AI Hardware
• Arm Unveils AI-Powered Technology to Enhance Mobile Device Graphics Quality
• Taiwan-Made Global Debut! HTC Launches AI Glasses VIVE Eagle: Features, Price, Release Date at a Glance
• Will AI Glasses Replicate iPhone's Explosive Growth Path? Tech Giants Vie for New Wearable Frontier

法院裁决动摇AI行业，版权问题引爆7500亿美元风险

简报：因在模型训练中涉嫌使用盗版书籍等未授权内容，OpenAI、Meta等AI巨头正面临法律诉讼，其中一份2025年6月的法院裁决表明，即使训练行为具“变革性”，使用盗版数据来源也无法规避侵权，预计Anthropic公司可能因此面临高达7500亿美元的赔偿。

相关链接：

• https://www.aibase.com/zh/news/20554

Court Ruling Rocks AI Industry, Copyright Issues Spark $750 Billion Risk 💥

Brief: AI giants such as OpenAI and Meta are currently embroiled in legal battles over allegations of using unauthorized content, including pirated books, for model training. A significant court ruling from June 2025 has clarified that even "transformative" training activities do not exempt companies from copyright infringement if pirated data sources are used. This decision could potentially expose Anthropic to damages totaling up to $750 billion. ⚖️💸

Related Links:

• https://www.aibase.com/zh/news/20554