Welcome to the "AI Daily" section! This is your guide to exploring the world of artificial intelligence every day. Every day, we present you with the latest content in the AI field, focusing on developers to help you understand technology trends and innovative AI product applications.
Fresh AI products click to learn more:https://app.aibase.com/zh
1. Microsoft launches its first self-developed image generation model MAI-Image-1, ranking in the top ten of LMArena
Microsoft AI has officially launched its first fully self-developed image generation model, MAI-Image-1, which ranks in the top ten of LMArena, demonstrating excellent image generation capabilities. MAI-Image-1 focuses on practical application value, avoiding repeated or monotonous image generation, while excelling in lighting effects and natural landscapes, combining speed and quality to improve creative efficiency. The Microsoft team has also tested the model on LMArena to ensure safe use and plans to apply it in Copilot and Bing Image Creator, providing users with more powerful creative tools.
【AiBase Summary:】
🌟 First self-developed model: MAI-Image-1 is Microsoft's first fully self-developed image generation model.
⚡ High quality and fast iteration: The model performs excellently in generating realistic images, combining speed and quality to improve creative efficiency.
🛡️ Safe and responsible application: Microsoft team tested the model on LMArena to ensure safe use and expects it to be available on more platforms.
Details link:https://microsoft.ai/news/introducing-mai-image-1-debuting-in-the-top-10-on-lmarena/
2. Baidu World 2025 will be held on November 13th: Focusing on large model technology, AI-native applications, and global strategy
Baidu World 2025 will focus on large model technology, AI-native applications, and global strategy, showcasing Baidu's technical advantages and future layout in the field of artificial intelligence.
【AiBase Summary:】
🧠 Baidu will showcase its latest progress and technical accumulation in the field of large model technology.
💡 Focus on building an AI-native application ecosystem, promoting digital transformation.
🌍 Showcasing Baidu's global expansion strategy, exploring potential for international market growth.
3. Meta's Super Intelligence Lab Launches New Technology, RAG Inference Speed Increased by 30 Times
Meta company established a new Super Intelligence Lab (MSL) and published its first important paper "REFRAG: Rethinking RAG based Decoding," significantly increasing the inference speed of large language models in retrieval augmented generation (RAG) tasks, with an increase of more than 30 times. This research improves the efficiency and accuracy of the model through information compression and continuous pre-training methods, injecting new momentum into the development of the artificial intelligence field.
【AiBase Summary:】
🧠 Meta established a Super Intelligence Lab, driving the development of AI technology.
⚡ The new paper "REFRAG" achieved a 30-fold increase in RAG inference speed, reducing computational requirements.
🚀 The REFRAG framework improves the efficiency and accuracy of large language models through information compression.
Details link:https://arxiv.org/abs/2509.01092
4. Build Your Own ChatGPT for $100: AI Expert Open-Sources nanochat Teaching Tool, Create a Chatbot in 4 Hours from Scratch
nanochat is a brand new open-source project that allows ordinary developers and AI enthusiasts to build a fully functional chat AI system at a very low cost. This project realizes a one-click process from data processing to deployment with a simple code stack, greatly reducing the technical barrier.
【AiBase Summary:】
🔥 nanochat provides a complete teaching tool to help users deeply understand the training process of large language models.
💡 The project covers model training, fine-tuning, evaluation, and interactive deployment, supporting distributed efficient loading.
🚀 Users just need to start a cloud node and run a single script to complete the entire process in 4 hours.
Details link:https://github.com/karpathy/nanochat
5. Google NotebookLM Launches Anime Video Feature: Nano Banana Generates Six Art Styles with One Click, Chinese Support Still Needs Optimization
Google's NotebookLM tool recently updated, adding an anime video feature. Using the Nano Banana image generation model, users can convert documents into lively videos and choose various art styles. Despite its powerful features, Chinese support still has shortcomings, such as dialect confusion and synchronization issues.
【AiBase Summary:】
✨ Google NotebookLM adds an anime video feature, enhancing the fun of content creation.
🎨 Supports six art styles, including watercolor, retro printing, and anime Japanese cartoon style.
⚠️ Chinese support has language processing issues and needs further optimization to adapt to multilingual environments.
6. China Agricultural University Launches Shennong Large Model 3.0, AI Helps Rural Revitalization
China Agricultural University launched Shennong Large Model 3.0 at the 2025 World Agricultural Science and Technology Innovation Conference. The model covers national agricultural disciplines and application scenarios, marking a new stage in China's agricultural AI technology. Shennong Large Model 3.0, by introducing 32B, 7B, and 1B versions, significantly improved computing efficiency and reduced computing power by 50%. Meanwhile, the team also launched the "Shennong Large Model Intelligent Body Platform," building a new AI application ecology, further promoting the development of agricultural AI technology towards a higher level of "system intelligence."
【AiBase Summary:】
🌱 Shennong Large Model 3.0 covers national agricultural disciplines and application scenarios, marking a new stage in China's agricultural AI technology.
🚀 Shennong Large Model 3.0 introduces 32B, 7B, and 1B versions, improving computing efficiency and reducing computing power by 50%.
🌍 The China Agricultural University team launched the "Shennong Large Model Intelligent Body Platform," building a new AI application ecology.
7. Tencent Focuses on AI Talent Development: Qingyun Scholarship First Period Funds 15 Excellent Master's and Doctoral Students, Provides Rare Computing Resources
Tencent announced the launch of the "Qingyun Scholarship," focusing on fundamental research and application innovation in the field of artificial intelligence, aiming to support outstanding master's and doctoral students in computer science, artificial intelligence, and their interdisciplinary fields. The first period plan will select 15 awardees, each receiving 500,000 RMB in cash and computing resources to solve computational bottlenecks in research. This scholarship is an important part of Tencent's talent program, further improving its comprehensive talent development system.
【AiBase Summary:】
🧠 Tencent's "Qingyun Scholarship" focuses on the field of artificial intelligence, supporting the research work of master's and doctoral students.
💰 The first period funds 15 awardees, each receiving 500,000 RMB in cash and computing resources.
🚀 The scholarship aims to improve Tencent's talent development system, providing mentorship, lab visits, and other opportunities.
8. Apple Launches New FS-DFM Model, AI Long Text Writing Efficiency Increased by 128 Times!
The FS-DFM model jointly released by Apple and the Ohio State University research team shows excellent performance in long text generation, requiring only 8 rapid iterations to generate high-quality text, with writing speed increased by up to 128 times, breaking previous efficiency bottlenecks in long text generation. The model optimizes the iteration mechanism through three steps, making it superior to other large models in key indicators such as perplexity and entropy in text generation, demonstrating its potential in the field of AI long text writing.
【AiBase Summary:】
🍎 The FS-DFM model requires only 8 iterations to generate quality comparable to traditional models with thousands of iterations.
🚀 Writing speed increased by up to 128 times, greatly improving the efficiency of long text generation.
🔍 Performance tests show that the FS-DFM outperforms other large models in key indicators such as perplexity and entropy.
Details link:https://machinelearning.apple.com/research/fs-dfm
9. Google Lens and AI Mode Launch Nano Banana Image Editing Function
Google officially launched the Nano Banana image editing function, which has received good feedback in the Gemini app. This function is integrated into AI mode and Google Lens, allowing users to edit and generate images more conveniently. Currently, the function is available to Android and iOS users in the United States and India, and is planned to be expanded to more countries and regions.
【AiBase Summary:】
📱 AI mode and Google Lens are integrated, enhancing the image capture and editing experience.
🌍 Added support for 35 languages and 40 countries, expanding the coverage of AI mode.
🎨 Users can create or edit images using the banana emoji symbol and have a Gemini watermark.
10. HKU and Meituan Jointly Solve AI Math Problems: CodePlot-CoT Lets Large Models Learn to Think with Code, Performance Surges 21%
The article introduces the CodePlot-CoT method proposed by the Hong Kong University and Meituan team, which enables large models to solve mathematical geometry problems by generating drawing code. This method transforms image generation problems into language modeling problems, improving the performance of large models in mathematical reasoning.
【AiBase Summary:】
📌 CodePlot-CoT solves math geometry problems by generating drawing code, enhancing reasoning ability.
💡 The Math-VR dataset requires models to actively draw and think, promoting the development of multimodal mathematical reasoning.
📈 Experimental results show that CodePlot-CoT's performance on the Math-VR benchmark test increased by up to 21%.
Details link:https://arxiv.org/pdf/2510.11718
