![[102dbe4d-dc57-405c-97f9-9b143be705e0.mp3]]

万兴天幕大模型文生视频能力位列国内第四

简报：

• 根据大模型评测平台SuperCLUE发布的《7月文生视频大模型全球榜单》，万兴科技旗下万兴天幕音视频多媒体大模型2.0的基模能力（基于华为盘古大模型）综合评估位列国内第四。
• 万兴科技与华为云已于7月29日举行“AI视频大模型实验室”揭牌仪式，双方将共同探索大模型创新及AI在数字创意场景的应用。

相关链接：

• 文生视频大模型全球榜发布，万兴科技旗下万兴天幕2.0基模能力跻身国内第四

Wondershare SkyView Large Model's Text-to-Video Capability Ranks Fourth Domestically

Briefing:

• Wondershare Technology's SkyView Audio-Video Multimedia Large Model 2.0's base model capability (based on Huawei's Pangu Large Model) has been comprehensively evaluated to rank fourth domestically, according to the "July Text-to-Video Large Model Global Ranking" released by the large model evaluation platform SuperCLUE. 🚀
• Wondershare Technology and Huawei Cloud held the unveiling ceremony for their "AI Video Large Model Lab" on July 29th, where both parties committed to jointly exploring large model innovation and the application of AI in digital creative scenarios. 💡

Related Links:

• Text-to-Video Large Model Global Ranking Released: Wondershare Technology's SkyView 2.0 Base Model Capability Enters Top Four Domestically 🔗

ChatGPT智能体成功绕过人机验证

简报：

• 近日有用户报告，ChatGPT智能体已能成功绕过Cloudflare的“我不是机器人”人机验证系统。
• 该智能体在执行任务时，能自动识别验证步骤，并模拟人类的鼠标移动轨迹点击“验证你是人类”的复选框，从而通过验证。
• Cloudflare的验证系统通常通过分析鼠标移动、点击速度等行为模式来区分人机，此次事件挑战了这类验证机制的有效性。
• 这一突破引发了业界对AI技术发展所带来的安全和隐私风险的担忧。

ChatGPT Agent Successfully Bypasses Human Verification

Brief:

• Recent user reports indicate that the ChatGPT agent has successfully bypassed Cloudflare's "I am not a robot" human verification system. 🤖
• When performing tasks, the agent can automatically identify the verification step, simulate human mouse movements, and click the "Verify you are human" checkbox to pass the verification.
• Cloudflare's verification system typically distinguishes between humans and bots by analyzing behavioral patterns such as mouse movements and click speed. This incident challenges the effectiveness of such verification mechanisms. 😬
• This breakthrough has raised concerns within the industry regarding the security and privacy risks posed by the advancement of AI technology. 🚨

三星HBM业务展望：HBM3E将主导下半年销售，HBM4样品已出

简报：三星电子表示，预计2025年下半年其HBM3E芯片的销售额将占公司HBM（高带宽内存）总销售额的90%以上，同时已开始向客户提供下一代HBM4芯片的样品。

相关链接：

• 三星电子：下半年HBM3E芯片销售将占逾90%

Samsung HBM Business Outlook: HBM3E to Dominate H2 Sales, HBM4 Samples Released 🚀

Briefing: Samsung Electronics announced it expects sales of its HBM3E chips to account for over 90% of the company's total HBM (High Bandwidth Memory) revenue in the second half of 2025. Additionally, the company has begun providing samples of its next-generation HBM4 chips to customers. 💡📈

Related Link:

• Samsung Electronics: HBM3E Chip Sales to Exceed 90% in Second Half

OpenAI 2023年收入破百亿，周活用户超7亿

简报：根据《The Information》报道，OpenAI在2023年商业化进程迅速，前七个月收入已达120亿美元，周活跃用户数突破7亿，并设定了到2029年年收入增至1250亿美元的目标。

相关连接：

• ​OpenAI 今年收入激增至 120 亿美元 周活跃用户破 7 亿大关

OpenAI's 2023 Revenue Exceeds $10 Billion, Weekly Active Users Surpass 700 Million! 🚀

Brief: According to reports from The Information, OpenAI experienced rapid commercialization in 2023. Its revenue reached an impressive $12 billion in the first seven months, and weekly active users have now surpassed 700 million. 📈 The company also aims to boost its annual revenue to$125 billion by 2029. 💰

Related Link:

• OpenAI's Revenue Surges to $12 Billion This Year, Weekly Active Users Break the 700 Million Mark

阿里开源WebShaper，GAIA得分超Claude 3.5与GPT-4o

简报：

• 阿里通义实验室发布了WebAgent系列的最新开源工具WebShaper，它采用创新的“形式化驱动”范式进行信息检索。
• 在专注于评估AI通用能力的GAIA基准测试中，基于WebShaper数据集训练的开源模型获得了60.19分，其表现超越了Claude 3.5 Sonnet和GPT-4o。
• WebShaper通过系统化的任务形式化方法生成训练数据，旨在确保知识结构与推理结构在语义上的一致性，以提升AI处理复杂任务的能力。
• 该项目已在GitHub开源，是阿里WebAgent生态的一部分，该生态还包括WebWalker、WebDancer等工具。

相关链接：

• 阿里WebShaper发布！GAIA碾压Claude 3.5 Sonnet、GPT-4o
• 项目地址

Alibaba Open-Sources WebShaper, GAIA Score Surpasses Claude 3.5 and GPT-4o 🚀

Briefing:

• Alibaba's Tongyi Lab has released WebShaper, the latest open-source tool in its WebAgent series, which utilizes an innovative "formalization-driven" paradigm for information retrieval.
• In the GAIA benchmark test, which focuses on evaluating AI's general capabilities, an open-source model trained on the WebShaper dataset achieved a score of 60.19, outperforming both Claude 3.5 Sonnet and GPT-4o. ✨
• WebShaper generates training data through a systematic task formalization method, aiming to ensure semantic consistency between knowledge and reasoning structures, thereby enhancing AI's ability to handle complex tasks.
• The project has been open-sourced on GitHub and is part of Alibaba's WebAgent ecosystem, which also includes tools like WebWalker and WebDancer. 💡

Related Links:

• Alibaba WebShaper Released! GAIA Overwhelms Claude 3.5 Sonnet, GPT-4o
• Project Address

我国大模型应用热度不减，注册用户总数破31亿

简报：

• 国家网信办消息，据不完全统计，我国大模型应用的个人用户注册总数已超过31亿。
• API调用用户总数超过1.59亿。
• 当前AI正通过网页、移动应用、API接口、本地部署、云服务部署等多种方式为用户提供服务。

相关链接：

• 我国大模型应用个人用户注册超31亿

China's Large Model Applications Maintain High Momentum, Total Registered Users Surpass 3.1 Billion 🔥

Briefing:

• According to the Cyberspace Administration of China (CAC), incomplete statistics show that the total number of individual user registrations for large model applications in China has exceeded 3.1 billion.
• The total number of API calling users has surpassed 159 million.
• Currently, AI 🤖 is providing services to users through various methods, including web pages, mobile applications, API interfaces, local deployment, and cloud service deployment. 📈

Related Links:

• China's Large Model Application Individual Users Registered Over 3.1 Billion