Welcome to the 【AI Daily】column! This is your daily guide to exploring the world of artificial intelligence. Every day, we present you with the hottest AI news, focusing on developers and helping you understand technology trends and innovative AI product applications.
Check out the latest AI products Learn More: https://top.aibase.com/
1. OpenAI Officially Releases GPT-4.5, Initially Available to ChatGPT Pro Users
OpenAI officially released its latest GPT-4.5 AI model on February 28th, marking a significant advancement in the field of conversational AI. GPT-4.5, through significantly expanding pre-training and post-training scales, improves unsupervised learning capabilities, exhibiting a more natural and fluent interactive experience and a deeper understanding of user intent. The new model excels in emotional intelligence, creativity, and reliability, making it widely applicable in writing, learning, and communication. It is expected to be initially available to ChatGPT Pro users.
【AiBase Summary:】
🌟 GPT-4.5 is OpenAI's largest and best-performing conversational AI model to date, boasting broader knowledge coverage and deeper understanding.
💡 The new model excels in emotional intelligence and creativity, helping to improve writing, coding, and problem-solving skills while reducing "hallucinations."
📈 GPT-4.5 utilizes new technology to train larger models using data from smaller models, widely applicable to intelligent agent planning, execution, and multi-step task processing.
2. Baidu's Ernie 4.5 Large Language Model to Launch on March 16th
Baidu announced that its latest Ernie 4.5 large language model will officially launch on March 16th. Since its release, the Ernie large language model has had a profound global impact. Version 4.5 significantly improves upon the capabilities of the base model, featuring advanced functionalities such as native multi-modality and deep thinking. Robin Li stated that open-sourcing this model aims to make it easier for users to experience its superior performance and promote its application in a wider range of fields, fully leveraging its potential value.
【AiBase Summary:】
🚀 Ernie 4.5, with native multi-modality and deep thinking capabilities, will launch on March 16th.
🔓 Baidu's decision to open-source Ernie 4.5 aims to make it easier for users to try and experience its performance.
🌍 Robin Li emphasizes that the true value of a foundational model lies in solving real-world problems and achieving large-scale applications.
3. Free Trial! Krea Launches Wan 2.1 Model: Stunning Dynamic Effects, Understands Complex Prompts
Krea officially launched its latest Wan 2.1 model, its most advanced AI video generation tool to date. The model is attracting attention for its exceptional dynamic effects and deep understanding of complex prompts. Users can now experience this new functionality for free, marking a significant breakthrough in AI video technology.
【AiBase Summary:】
🎥 The Wan 2.1 model is known for its outstanding video generation capabilities and understanding of complex prompts.
✨ User feedback suggests that Wan 2.1 is suitable for professional creators while also providing a convenient way for ordinary users to explore AI art.
🌐 Users can currently try Wan 2.1 for free via the official Krea website to experience this cutting-edge technology.
4. Pika 2.2 Officially Released: AI Video Generation Technology Receives a Major Upgrade
Pika Labs released Pika 2.2 on February 27th, 2025, bringing several significant feature upgrades, including longer video generation times and higher resolution. The new version's "Pikaframes" technology allows for smoother scene transitions, making it suitable for both casual and professional creators.
【AiBase Summary:】
🎥 Pika 2.2 supports video generation up to 10 seconds long, significantly increasing creative freedom.
🌟 Resolution has been increased to 1080p, ensuring the clarity and detail of the output video.
⚙️ The introduction of "Pikaframes" technology enables smoother scene transitions, expanding users' creative imagination.
5. DeepSeek Open Source Week Day Five: 6.6 TiB/s! 3FS Redefines AI Storage Infrastructure
On the final day of Open Source Week, DeepSeek released the high-performance parallel file system 3FS and the data processing framework Smallpond, marking a significant breakthrough in distributed storage technology. 3FS set a new industry record with a cluster throughput of 6.6 TiB/s, solving the data processing pain points of AI training and inference. Its decentralized architecture and strong consistency design improve storage performance, simplify the complexity of distributed development, and promote the democratization of AI infrastructure.
【AiBase Summary:】
⚡ 3FS achieves an aggregate read throughput of 6.6 TiB/s in an 180-node cluster through its decentralized architecture and strong consistency semantics; single-node KVCache lookup peaks exceed 40 GiB/s.
🔗 As the core of DeepSeek V3/R1, 3FS fully supports data preprocessing, checkpoint storage, and inference caching, simplifying the complexity of distributed development.
🌐 The open-sourcing of 3FS and Smallpond breaks through the storage bottlenecks of data-intensive applications and may lead to a significant improvement over traditional distributed systems.
Details: https://github.com/deepseek-ai/3FS
6. Free to Use! ByteDance's AI Programming Software Trae Integrates Claude 3.7, Developers Rejoice
ByteDance's AI programming software, Trae, has officially launched, integrating the latest Claude 3.7 model and offering unlimited free use, generating enthusiastic responses from developers. This move not only disrupts the traditional high-cost model of AI tools but also improves developer programming efficiency and experience. Claude 3.7's hybrid reasoning model excels in programming tasks, showcasing ByteDance's strong position in the AI programming field and potentially accelerating the popularization and application of AI programming.
【AiBase Summary:】
🚀 Trae integrates Claude 3.7, the world's first free open-source hybrid reasoning model, disrupting the traditional high-cost model.
💡 Claude 3.7 achieves a high score of 70.3% in programming tasks, demonstrating its advantages in rapid response and deep thinking.
🌍 ByteDance's free strategy may force competitors to reconsider their pricing models, promoting the popularization of AI programming tools.
7. Meta Plans to Launch a Standalone AI Chat App to Compete with ChatGPT
Meta is actively developing its standalone AI assistant app, Meta AI, aiming to compete with major market players such as OpenAI's ChatGPT and Google's Gemini. The app is expected to launch between April and June 2025 and will be available through Meta's social platforms and as a standalone app. Meta also plans to introduce a paid subscription service to offer more features and will hold its first AI developer conference, LlamaCon, at the end of April to showcase its latest technology and future plans, further solidifying its market position in the AI field.
【AiBase Summary:】
🌟 Meta plans to launch its standalone AI assistant app, Meta AI, between April and June 2025.
💰 Meta will test a paid subscription service to provide users with more features; the specific price has not yet been announced.
📅 Meta will hold its first AI developer conference, LlamaCon, at the end of April to showcase its latest technology.
8. Huawei's XiaoYi AI Assistant Web Version Launches, Supporting Q&A, Writing, and Programming
The Huawei XiaoYi family welcomes a new member: the XiaoYi web version, officially launched to provide users with comprehensive online AI services. Users can access it through a web browser and enjoy diverse functionalities, including Q&A, image recognition, reading, translation, and writing. First-time users need to register a Huawei account. After logging in, users can experience a clear layout and rich functional sections, especially the AI writing section, which provides professional writing assistance with a quick response to meet various user needs.
【AiBase Summary:】
💻 Users can access the Huawei XiaoYi web version through a web browser to enjoy diverse AI services.
📝 The AI writing section provides modules for promotional copywriting, market research, etc., offering professional writing assistance.
🔍 After logging in, users experience a clear layout, divided into four main sections: AI Q&A, writing, documents, and programming assistance.
Details: https://xiaoyi.huawei.com/
9. Image Generation Model Ideogram 2a Released: Double the Speed, Half the Cost, English Typography Rivals Human Design
On February 27th, 2025, Ideogram launched its new model, "Ideogram 2a," achieving significant improvements in generation speed and cost, further solidifying its leading position in the text-to-image field. Users highly praise the quality of the generated images, especially its excellent performance in English typography design, virtually flawless.
【AiBase Summary:】
🚀 The new model "Ideogram 2a" is online, significantly improving generation speed and reducing costs by half.
🎉 Ideogram excels in English typography design, generating image quality close to that of professional designers.
🌟 This update reflects Ideogram's continuous technological advancements, aiming to benefit more users with AI creation tools.
10. Shifting to AI Course Livestreaming! Actor Xie Mengwei ("Gazi Ge") Sells DeepSeek Course
Amidst the livestreaming craze, Xie Mengwei ("Gazi Ge"), the actor popular for his role in the TV series "Little Soldier Zhang Ga," has chosen the livestreaming path, launching an AI course called DeepSeek. This course claims that users can simply say "123123" into the camera to generate videos in multiple languages, seemingly very convenient. However, such claims have sparked many questions, with people curious about the actual effectiveness of the course and discussing whether they would be willing to purchase it for 10.9 yuan.
【AiBase Summary:】
📹 Xie Mengwei ("Gazi Ge") launches the DeepSeek AI course, attracting audience attention.
💰 The course costs only 10.9 yuan and claims simple operation, capable of generating videos in multiple languages.
❓ The effectiveness of the course is questioned, with many people discussing their willingness to purchase it online.
11. Honor and Alibaba Collaborate in the AI Field
Honor and Alibaba have partnered in the field of artificial intelligence to provide more intelligent services to global users. By integrating multiple advanced AI models, Honor's YOYO intelligent agent in the Magic7 series phones showcases richer AI functions. The new mode of YOYO assistant provides more accurate feedback, although slightly slower, improving the user experience.
【AiBase Summary:】
🤝 Honor and Alibaba collaborate, integrating AI models like Qianwen and Wanxiang to enhance user service experience.
📱 YOYO assistant launches the DeepSeek-R1 full-blooded version, supporting deep thinking mode and providing accurate feedback.
🌐 YOYO intelligent agent receives L3 certification in MagicOS 9.0, possessing 600 demand understanding and 900 task execution capabilities.
12. Lei Jun: Xiaomi's R&D Investment Projected to Reach 30 Billion Yuan in 2025, 105 Billion Yuan in the Next Five Years
At the Xiaomi 15 Ultra and Xiaomi SU7 Ultra product launch conference, CEO Lei Jun announced Xiaomi's massive investment in R&D. R&D investment is expected to exceed 24 billion yuan in 2024 and reach 30 billion yuan in 2025, with a quarter allocated to AI and related businesses. Over the next five years, Xiaomi's R&D investment will cumulatively reach 105 billion yuan, demonstrating Xiaomi's determination and strategic planning in technological innovation and market competition.
【AiBase Summary:】
💰 Xiaomi's R&D investment is projected to exceed 24 billion yuan in 2024.
🤖 A quarter of the 2025 R&D budget will be allocated to AI and related businesses.
📊 Xiaomi's cumulative R&D investment will reach 105 billion yuan over the next five years.
