Welcome to the AI Daily column! Your daily guide to exploring the world of artificial intelligence. We bring you the hottest AI news, focusing on developers and helping you understand technology trends and innovative AI product applications.
Discover new AI products Learn More: https://top.aibase.com/
1. Alibaba and Tencent Announce Full Support for the MCP Protocol
Recently, the Chinese AI field has seen a technological standard revolution, with the Model Context Protocol (MCP) becoming the de facto standard for the domestic AI ecosystem. Alibaba and Tencent's support marks a new round of competition among Chinese tech giants in the global AI standards race. As an open-source protocol, MCP simplifies the interaction between AI models and external tools, improving interoperability.
【AiBase Summary:】
🚀 The rapid rise of the MCP protocol highlights the urgent need for standardized protocols among Chinese tech companies.
🤝 Alibaba and Tencent's support will accelerate the adoption of MCP in China, promoting the application of AI agents.
⚖️ While the widespread adoption of MCP faces challenges, it could also give rise to a new AI development ecosystem, influencing future technological competition.
2. StepStar Launches New Multimodal Reasoning Model—Step-R1-V-Mini
StepStar's Step-R1-V-Mini is a groundbreaking multimodal reasoning model representing a significant advancement in the field. Supporting image and text input with text output, it boasts excellent instruction-following capabilities and versatility. Employing multimodal joint reinforcement learning and a verifiable reward mechanism, Step-R1-V-Mini excels in visual reasoning and mathematical logic, ranking among the top performers on the MathVision visual reasoning benchmark.
【AiBase Summary:】
🧠 Step-R1-V-Mini supports image and text input and text output, with strong instruction-following capabilities and versatility.
🔍 The model excels in visual reasoning, particularly ranking first domestically on the MathVision benchmark.
⚙️ It's available on the Step AI website and provides an API for developers and researchers.
Details: https://yuewen.cn/chats/new
3. Meitu WHEE Launches Image Generation Model Miracle F1
The WHEE platform recently launched Miracle F1, an AI image generation model that revolutionizes AI image creation with its exceptional image quality and deep understanding of complex concepts. It generates highly realistic images and excels in semantic understanding and stylistic diversity, catering to various user needs. Users can experience this visual magic through the official WHEE website.
【AiBase Summary:】
✨ Miracle F1 generates highly realistic images, simulating real-world lighting and material effects.
🧠 The model accurately understands complex concepts, improving creative efficiency and precision, almost like it has "mind-reading" capabilities.
🌈 Miracle F1 offers diverse styles, meeting the needs of e-commerce, event visuals, and illustrations.
4. Deep Research Now Powered by Gemini 2.5 Pro: Google's Smartest AI Model Takes Center Stage
Google announced an upgrade to its Deep Research feature with the experimental Gemini 2.5 Pro, showcasing exceptional reasoning capabilities and information integration. This breakthrough has garnered significant industry attention, marking a major milestone for AI research tools. Gemini 2.5 Pro not only improves search efficiency but also performs comprehensive analysis, transforming research methods and helping professionals adapt to new technologies. Google plans to expand Deep Research's applications to provide more intelligent support for academic and commercial research.
【AiBase Summary:】
🚀 The Gemini 2.5 Pro upgrade significantly improves Deep Research's search efficiency and analytical capabilities, handling complex topics and generating comprehensive reports.
📊 The model performs exceptionally well in various benchmark tests, particularly in long-context tasks with a context window of up to 1 million tokens, enabling analysis of massive datasets.
🌐 This technological advancement marks a significant milestone for AI research tools and is expected to revolutionize academic and commercial research.
5. Open-Source Model DeepCoder: Ultra-Efficient Programming, Surpassing OpenAI's o1 Model
The DeepCoder-14B-Preview model, jointly open-sourced by Together AI and Agentica, boasts 14 billion parameters and outperforms OpenAI's o1 model in programming tests. Its open-source content is comprehensive, including model weights, training data, and methods, facilitating in-depth research by developers. Through distributed reinforcement learning and high-quality datasets, DeepCoder demonstrates significant improvements in training efficiency and code quality, showcasing its immense potential in AI programming.
【AiBase Summary:】
🌟 The DeepCoder-14B-Preview model performs exceptionally well, surpassing OpenAI's o1 model.
📈 Comprehensive open-source content, including model weights and training data, facilitates developer research.
⚙️ Various techniques ensure data quality and training efficiency, significantly improving model performance.
Details: https://huggingface.co/agentica-org/DeepCoder-14B-Preview
6. Reasoning Performance Leaps Forward! DeepSeek Introduces Innovative SPCT Technology, Making Large Models More Empathetic
DeepSeek AI's Self-Play Principle Criticism Tuning (SPCT) technology marks a major breakthrough in large language models. This technology aims to build more general and scalable AI reward models, enhancing AI's understanding and response capabilities in complex environments. SPCT addresses challenges faced by existing reward models, such as input flexibility, accuracy, scalability during inference, and learning scalability, by dynamically generating principles and critiques.
【AiBase Summary:】
✨ SPCT technology aims to improve the generality and scalability of AI reward models, overcoming limitations of existing models.
💡 By dynamically generating principles and critiques, SPCT effectively improves AI performance and reasoning capabilities in complex tasks.
📈 DeepSeek-GRM-27B outperforms traditional models in several benchmark tests, demonstrating higher reward quality and scalability during inference.
Details: https://arxiv.org/abs/2504.02495
7. Anthropic Officially Releases! University Student Claude AI Usage Report Unveiled
This article explores the application of artificial intelligence (AI) in university student learning, specifically focusing on the use of Claude.ai. By analyzing a large amount of anonymized conversation data, the study reveals the usage preferences of students from different majors and the role of AI in learning. While AI offers convenience to students, it also raises concerns about outsourcing cognitive abilities, highlighting the challenges and opportunities facing educators in the AI era.
【AiBase Summary:】
📊 STEM students are early adopters of AI tools, with computer science students showing significantly higher usage rates than other majors.
🛠️ Students primarily use AI for creation and analysis, especially in designing educational content and solving technical problems.
🤔 AI usage raises concerns about outsourcing student cognitive abilities, and educators need to focus on balancing AI's supportive role with the development of students' fundamental skills.
8. Amazon Launches Next-Generation AI Voice Model Nova Sonic, Capturing Nuances in Tone, Intonation, and Rhythm
Amazon's newly launched AI voice model, Nova Sonic, aims to enhance the performance of its voice assistant, Alexa +. By processing voice locally, it generates natural and fluent responses, marking a significant breakthrough in speech recognition technology. Nova Sonic not only boasts speech recognition capabilities in complex environments but also adapts its responses based on user tone and style, improving user experience.
【AiBase Summary:】
🌟 Nova Sonic is Amazon's new AI voice model designed to enhance Alexa + performance.
💰 The model costs 80% less than OpenAI's GPT-4o, offering developers more choices.
🔊 Nova Sonic has speech recognition capabilities in complex environments, processing user requests quickly and accurately.
Details: https://www.aboutamazon.com/news/innovation-at-amazon/nova-sonic-voice-speech-foundation-model
9. Google NotebookLM to Launch Mobile App Version
Google's AI research tool, NotebookLM, is set to launch a standalone mobile client application, marking its expansion from web to mobile. This upgrade will provide users with a more convenient experience, meeting the demand for mobile applications. Since its launch, NotebookLM has garnered significant attention for its innovative features, and the future mobile application will further integrate Google's search capabilities, improving information processing efficiency.
【AiBase Summary:】
🚀 NotebookLM will launch on iOS and Android, improving mobile usability.
🔍 A new "Discover Sources" feature allows users to automatically search and integrate web content into their notebooks.
🎙️ Future integration with Google Search may enable conversion from URLs to summaries and mind maps.
10. AI Video Generation Technology TTT: Directly Outputs One-Minute Complete Tom and Jerry Animation Without Editing or Splicing
This research, by introducing a test-time training layer, successfully generated a one-minute Tom and Jerry animation video, marking a new breakthrough in AI video generation technology. The technology excels in visual coherence and narrative integrity, requiring no post-production editing, demonstrating AI's immense potential in creative content production. Despite some imperfections, its application prospects are vast, and it is expected to change video production methods in the future.
【AiBase Summary:】
🚀 By introducing a TTT layer, the model can generate a complete one-minute animation without post-editing.
🎨 The generated video excels in temporal consistency and narrative coherence, approaching the quality of traditional animation.
💡 This technology is expected to reduce video production costs, accelerate creative workflows, and be scalable to more complex content in the future.
Details: https://test-time-training.github.io/video-dit/
11. Cyberspace Administration of China: 346 Generative AI Services Completed Registration as of March 31, 2025
On April 8, the Cyberspace Administration of Shanghai released an announcement detailing the registration status of generative AI services as of March 31, 2025. In accordance with the requirements of the Cyberspace Administration of China, relevant departments jointly promoted the registration of generative AI services to foster innovation and regulate applications in this field.
【AiBase Summary:】
🌟 As of March 31, 2025, 346 generative AI services have completed registration with the Cyberspace Administration of China.
📊 159 generative AI applications accessed via APIs have been registered with local Cyberspace Administration offices.
🔍 All online applications must publicly disclose information about the registered services used, including model names and registration numbers.
