![[99f045de-7644-4c9f-969e-a392fd784c66.mp3]]

Meta加速AI商业化，WhatsApp商务套件推AI语音功能

简报：

• Meta公司宣布将为WhatsApp商务套件引入AI语音通话功能，面向大型企业开放，探索AI语音代理等应用场景。
• 此举是Meta加快AI技术落地和商业化的重要举措，旨在提升企业客户的自动化和服务能力。
• 同步，Meta持续整合AI资源，并加大在人工智能领域的投入和研发布局。

相关链接：

• Meta探索AI商品推荐 - 36氪
• Meta推进AI语音助手应用 - aibase

🚀 # Meta Speeds Up AI Commercialization, WhatsApp Business Suite Introduces AI Voice Features 📞

Briefing:

• Meta Platforms announced it will introduce AI voice calling features to the WhatsApp Business Suite, making them available to large enterprises and exploring applications like AI voice agents.
• This move is a significant step for Meta to accelerate the implementation and commercialization of AI technology, aiming to enhance enterprise clients' automation and service capabilities.
• Concurrently, Meta continues to integrate AI resources and increase its investment and R&D layout in artificial intelligence. 🤖

Related Links:

• Meta Explores AI Product Recommendations - 36Kr
• Meta Advances AI Voice Assistant Applications - aibase

Gemini 2.5 Pro引领AI大模型新阶段

简报：

• 谷歌DeepMind最新发布的Gemini 2.5 Pro大语言模型，凭借百万级长上下文处理能力和更高效的信息检索方式，引发业界高度关注，被认为将推动AI应用进入革命性新阶段。
• Gemini 2.5 Pro在AI编程、信息检索等领域表现突出，能够一次读取完整项目内容，提升用户体验，并结合RAG（检索增强生成）技术，实现更高的信息召回率。
• 谷歌DeepMind科学家指出，长上下文和用户即时输入使模型更加个性化和时效，预计未来千万级上下文处理将成行业标准，为各类AI应用带来突破。

相关链接：

• Gemini2.5Pro助推大模型革命
• Google's Gemini 2.5 Pro is the smartest model you're not using - and 4 reasons it matters for enterprise AI - VentureBeat
• Gemini 2.5 Pro is now available without limits and for cheaper than Claude, GPT-4o - VentureBeat

Gemini 2.5 Pro Ushers in a New Era for Large AI Models

Briefing:

• Google DeepMind's newly released Gemini 2.5 Pro large language model has garnered significant industry attention 🤩, thanks to its million-token long-context processing capabilities and more efficient information retrieval methods. It is expected to propel AI applications into a revolutionary new phase.
• Gemini 2.5 Pro excels in areas such as AI programming and information retrieval. It can read entire project contents in one go, significantly enhancing user experience, and integrates with RAG (Retrieval-Augmented Generation) technology to achieve higher information recall rates.
• Scientists at Google DeepMind highlight that long contexts and real-time user input make the model more personalized and current. They anticipate that processing tens of millions of tokens will become the industry standard in the future, bringing breakthroughs across various AI applications. 🚀

Related Links:

• Gemini 2.5 Pro Propels Large Model Revolution
• Google's Gemini 2.5 Pro is the smartest model you're not using - and 4 reasons it matters for enterprise AI - VentureBeat
• Gemini 2.5 Pro is now available without limits and for cheaper than Claude, GPT-4o - VentureBeat

淘宝推百亿参数推荐大模型RecGPT，升级个性化购物体验

简报：

• 7月1日，淘天集团于“硬核少年技术节4.0”正式发布自产百亿参数推荐大模型RecGPT，应用于淘宝首页“猜你喜欢”，实现生成式推荐（AIGR）技术升级。
• 新系统带来用户点击量两位数增长，同时加购和页面停留时长提升超5%，显著增强了个性化推荐的准确性和互动体验。
• RecGPT基于淘宝星辰LLM大模型，通过强化学习深入训练十余年用户消费行为，结合多模态认知整合数亿商品图文信息，并能自动生成个性化推荐理由和文案。
• 升级使淘宝和天猫在索引、推荐等多个业务线部署大模型，推动电商平台技术和个性化体验创新。

相关链接：

• 淘宝全新推荐大模型 RecGPT 上线，购物体验再升级

Taobao Launches Billion-Parameter Recommendation Large Model RecGPT, Upgrading Personalized Shopping Experience

Briefing:

• On July 1st, Taotian Group officially launched its self-developed billion-parameter recommendation large model, RecGPT, at the "Hardcore Youth Tech Festival 4.0". It has been applied to Taobao's homepage 'Guess You Like' section, marking an upgrade in Generative Recommendation (AIGR) technology. ✨
• The new system has led to double-digit growth in user click-through rates, alongside an over 5% increase in add-to-cart rates and page dwell time. This significantly enhances the accuracy of personalized recommendations and improves the interactive experience. 📈
• RecGPT is built upon Taobao's Xinchen LLM large model. It has been extensively trained using reinforcement learning on over ten years of user consumption behavior, integrating hundreds of millions of product image and text information through multi-modal cognition. It can also automatically generate personalized recommendation reasons and copy. 🤖
• This upgrade enables Taobao and Tmall to deploy large models across various business lines, including indexing and recommendations, thereby driving technological advancement and personalized experience innovation in e-commerce platforms.

Related Link:

• Taobao's new recommendation large model RecGPT launched, shopping experience further upgraded

英特尔与百度合作推动文心大模型4.5端侧部署

简报：

• 6月30日，百度正式开源了文心大模型4.5系列，发布十款新模型并提供权重和推理代码，方便开发者免费使用；
• 英特尔凭借OpenVINO工具套件，在发布当天即完成对文心4.5端侧模型的适配，并实现在酷睿Ultra平台的端侧部署；
• 百度飞桨团队与英特尔OpenVINO团队自2021年起深度合作，推动包括文心4.5等多款AI模型的优化部署；
• 本次合作展示了AI模型端侧应用加速和开放生态的最新进展。

相关链接：

• AIBase 新闻
• AIBase 新闻2
• IT之家报道

Intel and Baidu Partner to Drive Wenxin Large Model 4.5 Edge Deployment 🤖

Briefing:

• On June 30th, Baidu officially open-sourced its Wenxin Large Model 4.5 series, releasing ten new models and providing weights and inference code for developers to use for free.
• Leveraging its OpenVINO toolkit ✨, Intel completed the adaptation of Wenxin 4.5 edge models on the very day of their release, successfully deploying them on the Core Ultra platform.
• Baidu's PaddlePaddle team and Intel's OpenVINO team have been collaborating closely since 2021, driving the optimized deployment of various AI models, including Wenxin 4.5.
• This collaboration showcases the latest advancements in accelerating AI model edge applications and fostering an open ecosystem. 🚀

Related Links:

• AIBase News
• AIBase News 2
• IT之家 Report

微软新AI系统在复杂医疗诊断中准确率超医生四倍

简报：

• 微软近日发布了AI诊断系统MAI-DxO，在复杂医疗病例测试中诊断准确率为79.9%，是经验丰富医生的四倍，同时将医疗成本降低近70%。
• 该系统通过模拟临床逐步决策流程，与21名来自美国和英国的医生在304个真实复杂病例中对比，医生准确率为19.9%，平均成本为2963美元，而MAI-DxO准确率为79.9%，成本降至2397美元。
• MAI-DxO采用多虚拟医生协同诊断机制，显著提升了诊断性能和成本效益，但研究团队指出其测试局限于复杂教学案例，结果可能不代表日常常见病例。

相关链接：

• 微软AI医疗系统四倍超越医生，成本下降近七成
• Microsoft MAI-DxO AI beats 21 doctors in medical diagnosis test, scores 85.5% accuracy
• Microsoft's medical AI system is four times more accurate than human doctors: Here's how

Microsoft's New AI System Achieves Four Times Higher Accuracy Than Doctors in Complex Medical Diagnoses 🚀

Brief:

• Microsoft recently unveiled its AI diagnostic system, MAI-DxO, which achieved a diagnostic accuracy of 79.9% in complex medical case tests, four times higher than experienced doctors, while simultaneously reducing medical costs by nearly 70%. 😮
• The system simulates a step-by-step clinical decision-making process. In a comparison with 21 doctors from the US and UK involving 304 real complex cases, doctors achieved an accuracy of 19.9% with an average cost of $2963, whereas MAI-DxO reached 79.9$2397.
• MAI-DxO employs a collaborative diagnostic mechanism involving multiple virtual doctors, significantly enhancing diagnostic performance and cost-effectiveness. However, the research team noted that its testing was limited to complex teaching cases, meaning the results may not be representative of common daily cases. 🩺

Related Links:

• Microsoft AI Medical System Four Times More Accurate Than Doctors, Costs Down Nearly 70%
• Microsoft MAI-DxO AI beats 21 doctors in medical diagnosis test, scores 85.5% accuracy
• Microsoft's medical AI system is four times more accurate than human doctors: Here's how