Scroller Briefings - Ant Group Open-Sources Ring-lite Lightweight Inference Model, Innovative Reinforcement Learning Method Boosts Performance Across Multiple Domains 🧠
编辑
目录
![[5ac1f34b-b42d-4586-bc04-d0299e73077f.mp3]]
蚂蚁开源Ring-lite轻量级推理模型，创新强化学习方法助力多领域表现
简报：
- 蚂蚁技术团队近期正式开源了轻量级MoE推理模型Ring-lite，模型总参数16.8B，激活参数2.75B，基于自主研发的C3PO强化学习训练方法，显著提升了推理性能。
- Ring-lite在AIME24/25、LiveCodeBench、CodeForce、GPQA-diamond等多项推理榜单成绩优异，数学高考测试可达130分左右，性能领先同级别模型。
- 团队不仅开放了模型权重和训练代码，还计划逐步公开全部数据集、超参配置与实验记录，实现全链路透明化，推动轻量级MoE模型研究。
相关链接：
Ant Group Open-Sources Ring-lite Lightweight Inference Model, Innovative Reinforcement Learning Method Boosts Performance Across Multiple Domains 🧠
Brief:
- Ant Group's technical team recently open-sourced Ring-lite, a lightweight MoE (Mixture of Experts) inference model. The model boasts 16.8B total parameters with 2.75B active parameters and significantly enhances inference performance using their self-developed C3PO reinforcement learning training method.
- Ring-lite has achieved excellent results on multiple inference benchmarks, including AIME24/25, LiveCodeBench, CodeForce, and GPQA-diamond. It scores approximately 130 points on college entrance exam-level math tests, outperforming models in its class. 🚀
- The team has not only released the model weights and training code but also plans to progressively disclose all datasets, hyperparameter configurations, and experimental records to achieve full-chain transparency and advance lightweight MoE model research. 💡
Related Links:
开心消消乐接入鸿蒙生态后日变现流水实现近百倍增长
简报：
- 在HDC 2025鸿蒙生态增长峰会上，开发者开心消消乐宣布自2024年12月接入鲸鸿动能变现服务、结合端云协同精准推荐及开发者激励计划后，其游戏的日变现流水实现了近百倍增长，流水规模持续攀升。
相关链接：
Happy Match Sees Nearly Hundredfold Daily Revenue Increase After Integrating with HarmonyOS Ecosystem 🥳
Brief:
- At the HDC 2025 HarmonyOS Ecosystem Growth Summit, developer Happy Match announced that since integrating with Petal Ads monetization services in December 2024, combined with device-cloud collaborative precise recommendations and the developer incentive program, its game's daily monetization revenue has achieved a nearly hundredfold increase, with revenue scale continuing to climb. 🚀💰
Related links:
ChatGPT 推出 macOS 隐形会议笔记工具，支持智能转录与结构化整理
简报：
- OpenAI 发布了专为会议、头脑风暴和个人管理设计的 ChatGPT 新笔记工具，目前仅限 macOS 桌面端使用。
- 用户可通过“隐形”录音，工具会自动转录和实时分析会议内容，生成结构化笔记并保存于 Canvas 界面，无需额外机器人介入会议。
- 笔记功能可通过提示词自动提炼会议决策、总结讨论等关键信息，还结合记忆功能实现个性化格式调整。
- 该工具已于2025年6月4日面向 Team 用户开放，6月18日扩展至 Pro、Enterprise 和 Edu 用户，未来有望支持更多平台。
- 市场整体反馈积极，但与 Notion、Granola 等竞品相比，协作和多平台支持仍待提升。
相关链接：
ChatGPT Launches Invisible Meeting Notes Tool for macOS, Featuring Smart Transcription and Structured Organization
Brief: 🚀
- OpenAI has released a new ChatGPT note-taking tool designed for meetings, brainstorming, and personal management, currently exclusively available for macOS desktop.
- Users can utilize "invisible" recording, allowing the tool to automatically transcribe and analyze meeting content in real-time, generating structured notes saved within a Canvas interface, without the need for additional bots to join the meeting.
- The note-taking feature can automatically extract key information such as meeting decisions and discussion summaries via prompts, and also incorporates a memory function for personalized formatting adjustments.
- The tool was launched for Team users on June 4, 2025, and will be extended to Pro, Enterprise, and Edu users on June 18, with future support for more platforms anticipated.
- Overall market feedback is positive, though collaboration and multi-platform support still need improvement compared to competitors like Notion and Granola. 📝
Related Links: 🔗
MiniMax Hailuo 02在用户基准测试中性能领先谷歌Veo 3
简报：
- MiniMax 发布的第二代视频AI模型 Hailuo 02，在人工分析视频竞赛（Artificial Analysis Video Arena）基准测试中，在图像到视频类别获得第二名，仅次于字节跳动的 Seedance，并超越了谷歌的 Veo 3。
- Hailuo 02采用噪声感知计算重分配（NCR）架构，训练和推理效率提升2.5倍，参数量和训练数据量较前代产品大幅增加，并提升了模拟复杂场景（如体操动作）的能力。
- Hailuo 02提供更高分辨率（最高1080p）和更长时长（最多十秒）的多种视频生成变体，用户生成视频的成本远低于谷歌Veo 3。
- 自2023年8月推出以来，Hailuo平台共生成超37亿条视频，受到全球创作者广泛关注。
相关链接：
MiniMax Hailuo 02 Outperforms Google Veo 3 in User Benchmarks
Brief:
- MiniMax's second-generation video AI model, Hailuo 02, secured second place in the Image-to-Video category of the Artificial Analysis Video Arena benchmark test, trailing only ByteDance's Seedance and surpassing Google's Veo 3.
- Hailuo 02 employs a Noise-Aware Compute Redistribution (NCR) architecture, boosting training and inference efficiency by 2.5 times. It also features a significant increase in parameter count and training data compared to its predecessor, enhancing its ability to simulate complex scenarios like gymnastic movements. 💪
- Hailuo 02 offers various video generation variants with higher resolution (up to 1080p) and longer durations (up to ten seconds). The cost for users to generate videos is significantly lower than Google Veo 3. 💰
- Since its launch in August 2023, the Hailuo platform has generated over 3.7 billion videos, attracting widespread attention from creators globally. 🌍
Related Links:
- MiniMax Hailuo 02 User Benchmark Exceeds Google Veo 3
- MiniMax New Video AI Model Technology Interpretation
- Hailuo 02 Release and Industry Impact
腾讯云发布AI Builder平台，普通用户可用自然语言开发应用
简报：
- 腾讯云日前正式推出国内首个全链路AI驱动应用开发平台“AI Builder”，通过自然语言处理技术显著降低开发门槛，无需编程经验即可快速生成小程序或网页应用。
- 用户只需描述需求，AI Builder便能自动完成从页面设计到后端配置和上线全流程，支持一键部署并提供可视化修改和导出完整代码包功能。
- 该平台旨在让更多非技术背景人士参与数字应用开发，提升创意实现效率。
相关链接：
Tencent Cloud Launches AI Builder Platform, Enabling Ordinary Users to Develop Applications with Natural Language
Briefing:
- Tencent Cloud recently officially launched 'AI Builder', the first full-stack AI-driven application development platform in China. It significantly lowers the development barrier through natural language processing technology, allowing users to quickly generate mini-programs or web applications without any programming experience. 🚀
- Users only need to describe their requirements, and AI Builder automatically handles the entire process from page design to backend configuration and online deployment. It supports one-click deployment and offers functions for visual modification and exporting complete code packages. ✨
- The platform aims to empower more non-technical individuals to participate in digital application development, enhancing the efficiency of bringing creative ideas to life. 💡
Related Links:
如果对你有用的话，可以打赏哦
打赏
本文作者:topwind
本文链接:
版权声明:本博客所有文章除特别声明外，均采用 BY-NC-SA 许可协议。转载请注明出处！
目录