Welcome to the 【AI Daily】 section! This is your daily guide to exploring the world of artificial intelligence. Every day, we present you with hot topics in the AI field, focusing on developers to help you gain insights into technology trends and innovative AI product applications.
Fresh AI products Click to Learn More: https://top.aibase.com/
1. The official version of "Jichuang," the AIGC tool under ByteDance, is now online
The AIGC tool "Jichuang" from ByteDance has officially launched, offering various content generation methods, including AI script generation, intelligent video production, and digital humans. The platform utilizes deep learning and natural language processing, allowing users to generate high-quality text and video content simply by inputting a topic. The newly added viral split and one-click review features help creators respond quickly to market changes, improving creation efficiency and content compliance.
【AiBase Highlights:】
🌟 Jichuang supports multiple content generation methods, enhancing creation efficiency.
📊 New viral split and one-click review features help creators quickly respond to the market.
🎨 AI technology leads the content creation revolution, encouraging creators to explore new tools.
Details link: https://top.aibase.com/tool/douyinjichuang
2. Tongyi Wansiang's Video Model 2.1 Significantly Enhances Complex Motion Capabilities
Alibaba's Tongyi Wansiang video generation model has undergone a major upgrade in the latest version 2.1, launching both a fast version and a professional version, focusing on efficient performance and outstanding expressiveness. This version has made significant progress in handling complex motions and restoring real physical laws, enhancing the cinematic quality of videos and adherence to instructions.
【AiBase Highlights:】
🚀 The new version includes a fast version and a professional version, enhancing efficient performance and expressiveness.
🐱 The generated videos are rich in detail, successfully solving the "ghost drawing" problem, achieving precise generation of Chinese and English text.
🎬 Camera movement effects resemble those of a film master, automatically adjusting shots to enhance video artistry.
Details link: https://tongyi.aliyun.com/wanxiang/videoCreation
3. Microsoft Officially Open-Sources the Powerful Small Model Phi-4
Microsoft has released the small language model Phi-4 on the Hugging Face platform, with 14 billion parameters, performing exceptionally well and surpassing several well-known models such as GPT-4o and Llama-3.1. Phi-4 scored 91.8 in the AMC math competition and 84.8 in the MMLU test, demonstrating strong reasoning abilities. The model uses an innovative synthetic data generation method and supports a context length of up to 16k, suitable for consumer-grade hardware.
【AiBase Highlights:】
🧩 The Phi-4 model launched by Microsoft has only 14 billion parameters but outperforms multiple well-known models in performance tests.
📈 Phi-4 excels in math and reasoning abilities, particularly achieving high scores in AMC and MMLU tests.
🌍 The model is open-sourced and supports commercial use, attracting attention from many developers and AI enthusiasts.
Details link: https://huggingface.co/microsoft/phi-4
4. New Video Restoration Technology SeedVR: Blurry to HD, Can Handle Videos of Any Length
In the context of rapid digital media development, improving video quality has become an important issue. The SeedVR technology developed by the research team from Nanyang Technological University and ByteDance utilizes an innovative moving window attention mechanism to significantly improve video restoration effects, especially when processing AI-generated videos. This technology can handle videos of any length and effectively restore image details, providing users with a more realistic visual experience.
【AiBase Highlights:】
🌟 SeedVR utilizes a moving window attention mechanism, successfully enhancing the processing capability for long video sequences.
🎥 The technology uses larger window sizes, significantly improving the restoration quality of high-resolution videos.
🚀 Combining various modern technical means, SeedVR performs excellently in multiple benchmark tests, especially suitable for AI-generated videos.
Details link: https://iceclear.github.io/projects/seedvr/
5. Adobe's TransPixar Seamlessly Integrates Smoke, Reflections, and Other Transparent Effects into Scenes
TransPixar, a system co-developed by Adobe Research and the Hong Kong University of Science and Technology, revolutionizes visual effects production, especially in handling transparent elements. This technology generates visual effects that include an alpha channel, significantly improving production efficiency and reducing costs. The launch of TransPixar coincides with the industry's surge in demand for high-quality effects, signaling a transformation in future film production processes.
【AiBase Highlights:】
🌟 TransPixar can generate high-quality transparent effects with limited training data through innovative AI technology, simplifying the special effects production process.
💡 This system not only enhances the work efficiency of large production teams but also reduces production costs for small studios, enabling them to achieve complex effects.
🚀 TransPixar shows great potential in real-time applications, quickly generating transparent effects suitable for video games and augmented reality scenarios.
6. ByteDance Teams Up with Universities to Launch STAR Model: Enhancing Video Clarity and Resolution
The STAR technology, jointly developed by a research team from Nanjing University, ByteDance, and Southwest University, achieves video super-resolution processing through a text-to-video model, significantly enhancing the quality of low-resolution videos. This technology combines spatiotemporal enhancement methods and is suitable for low-clarity videos downloaded from video-sharing platforms. The research team has released pre-trained models and inference code on GitHub, making the usage process straightforward and advancing the field of video processing.
【AiBase Highlights:】
🌟 The new STAR technology combines text-to-video models to achieve video super-resolution, enhancing video quality.
🛠️ The research team has released pre-trained models and inference code, making the usage process simple and clear.
📩 Contact information is provided to encourage users to communicate and discuss with the research team.
Details link: https://github.com/NJU-PCALab/STAR
7. Stability AI Launches SPAR3D: Generate 3D Objects from a Single Image in One Second
At CES, Stability AI introduced SPAR3D, an innovative two-stage 3D generation technology that can create precise 3D objects from a single image in less than a second. SPAR3D not only provides an efficient way to design 3D prototypes but also allows users to edit the generated 3D objects in real-time, enhancing creative flexibility.
【AiBase Highlights:】
⚡ Real-time generation capability: SPAR3D can generate a complete 3D object from a single image in one second and supports quick editing.
🔍 Accurate structural prediction: Provides precise geometric shapes and 360-degree views, including hidden areas, ensuring high-resolution output.
🚀 Open usage policy: Supports both commercial and non-commercial use, allowing users to access and utilize the model in various ways.
Details link: https://stability.ai/news/stable-point-aware-3d
8. 238 Generative AI Services Registered with the National Cyberspace Administration by 2024
On January 8, the National Internet Information Office announced that by December 31, 2024, 302 services had completed registration, with 238 newly added in 2024. This indicates rapid development in the field, with more new services gaining official recognition. The announcement also requires companies providing opinion-related services to register and enhance the transparency of already launched applications to ensure compliance and safety.
【AiBase Highlights:】
🌟 In 2024, a total of 302 generative AI services completed registration with the National Cyberspace Administration, with 238 newly added.
🔍 Companies providing opinion-related services can register through local cyberspace departments to ensure compliance.
📝 Launched applications must disclose the registered service information used to enhance transparency.
9. 2024 Hurun Top 50 AI Companies in China: iFlytek Second, SenseTime Third
The 2024 Hurun Top 50 AI Companies in China list has been officially released, with Cambricon ranking first with a valuation of 238 billion RMB, showcasing its strong capabilities in the AI field. iFlytek and SenseTime follow closely with valuations of 116 billion and 50 billion respectively. The list shows that Beijing, Shanghai, and Shenzhen are the main hubs for AI companies, accounting for 80% of the listed firms.
【AiBase Highlights:】
🌟 Cambricon ranks first with a valuation of 238 billion RMB, with iFlytek and SenseTime coming in second and third.
🏙️ Beijing, Shanghai, and Shenzhen have become the main hubs for AI companies, collectively accounting for 80% of the listed firms.
📊 This list focuses on non-embodied intelligence companies, excluding companies in robotics and smart home fields.
10. You Wouldn't Believe It! Repeat This Simple Command to Make AI Generate Code 100 Times Faster
Max Woolf, a senior data scientist at BuzzFeed, conducted an experiment using the Claude 3.5 language model. By repeatedly asking the AI to improve the code, he successfully reduced the original run time from 657 milliseconds to 6 milliseconds, achieving a 100-fold performance increase. The AI demonstrated a unique understanding of "better code" by automatically adding enterprise features.
【AiBase Highlights:】
🌟 AI enhances code performance through repeated commands, reducing the original code run time from 657 milliseconds to 6 milliseconds.
💡 AI automatically adds enterprise features to the code, showcasing its unique understanding of "better code."
🛠️ Prompt engineering remains important; precise requests can accelerate result generation, but manual developers still need to verify and fix the outputs.
11. NVIDIA Launches Groot Teleop Technology Allowing Robot Training via Apple Vision Pro
NVIDIA introduced a series of innovative technologies at CES aimed at accelerating the development of humanoid robots, particularly in industrial and manufacturing fields. Through the Isaac GR00T blueprint, developers can use imitation learning to generate a large amount of synthetic motion data to train humanoid robots. This technology not only reduces the time and cost of data collection but also generates physical awareness videos through the Cosmos platform, promoting the development of physical AI.
【AiBase Highlights:】
🚀 NVIDIA launched the Isaac GR00T blueprint, using imitation learning to generate synthetic motion data, accelerating humanoid robot development.
📊 The Cosmos platform has been trained on 18 trillion data points, generating physical awareness videos and promoting the development of physical AI.
🤝 Several robotics companies have adopted Isaac GR00T technology, showing good application results.
12. Shocking Invention! Engineer Creates Robot Rifle Using ChatGPT, OpenAI Responds Immediately
Recently, engineer STS3D created a robot rifle that can aim and shoot based on ChatGPT commands, showcasing impressive response speed and accuracy. This invention has sparked widespread discussion, especially concerning the concerns of turning sci-fi technology into reality. OpenAI quickly responded, pointing out that this behavior violates company policy, prohibiting the use of its services to develop weapons.
【AiBase Highlights:】
🌐 OpenAI swiftly severed ties with the engineer developing the robot rifle due to violations of usage policies.
🤖 The STS3D robot rifle can aim and shoot based on ChatGPT commands, demonstrating high accuracy.
⚔️ Although OpenAI revised its policy last year, it still prohibits the use of its services to develop any form of weapons.
13. Due to User Complaints, Microsoft Rolls Back Bing AI Image Generator Upgrade, Returns to Old DALL-E
Recently, Microsoft decided to roll back to the old DALL-E model due to user dissatisfaction with the new version of the Bing image generator. User feedback indicated that the new version significantly degraded image quality and detail handling, leading Microsoft’s search department head Jordi Ribas to acknowledge the issue and take steps to restore the old version.
【AiBase Highlights:】
🌟 Microsoft decided to roll back the new version of the Bing image generator due to user complaints, aiming to improve user experience.
🎨 The upgraded image generation effects did not meet expectations, with users reporting a noticeable decline in detail and quality.
🔧 Jordi Ribas stated that they will revert to the old DALL-E model, with adjustments expected to take a few weeks to complete.