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 technological trends and understand innovative AI product applications.
Fresh AI products click to learn more: https://top.aibase.com/
1. Kling AI API's Lip Syncing Capability Fully Opened, Virtual Try-On Upgraded to V1.5 Model
Beijing Kuaishou Technology Co., Ltd. recently announced that the Kling AI API has completed a new round of upgrades, achieving significant progress in virtual try-on and lip-syncing functionalities. The upgraded V1.5 model supports "top + bottom" clothing combinations, enhancing the realism of the try-on experience. Additionally, the full opening of lip-syncing capabilities makes the generated video content more vivid, bringing new creative possibilities. These technological advancements will provide strong support for clients in e-commerce, advertising, and marketing, promoting innovation and development in visual content.
【AiBase Summary:】
👗 The Kling AI V1.5 model supports combined clothing, enhancing the realism and practicality of virtual try-ons.
🎤 The lip-syncing capability is fully opened, achieving perfect synchronization between video characters' lip movements and voiceovers.
🚀 This upgrade will help enterprise users take new steps in business growth and promote visual content innovation.
2. Doubao Large Model Claims to Have Caught Up with GPT-4, Reveals 3 Million Long Text Capability for the First Time
ByteDance's Doubao large model announced in its 2024 technical progress report that its latest version, Doubao-pro-1215, has fully aligned with GPT-4 in comprehensive performance and excels in certain professional fields. This progress marks the rise of China's large model technology, demonstrating significant improvements in understanding accuracy and generation quality, especially surpassing GPT-4 in complex scenarios while offering more competitive service prices.
【AiBase Summary:】
🚀 The Doubao large model has fully aligned with GPT-4 in comprehensive performance and performs better in some professional fields.
💡 Through optimizing data processing and innovating model architecture, Doubao has made significant progress in understanding accuracy and generation quality.
📚 For the first time, it publicly disclosed the ability to handle 3 million characters of ultra-long text, with processing delays controlled within 15 seconds.
3. Zhiyuan's Deep Reasoning Model GLM-Zero Preview Version Launched
Zhiyuan Huazhang Technology Co., Ltd. released the first version of its reasoning model GLM-Zero-Preview, based on extended reinforcement learning technology, at the end of the year. This model focuses on enhancing AI's reasoning abilities in areas like mathematical logic and code writing, showing excellent performance. Although there is still a gap compared to OpenAI's models, the company plans to continue optimizing and expanding its application areas. Users can experience this model on the Zhiyuan Qingtalk platform, and developers can also access it through API calls.
【AiBase Summary:】
🚀 GLM-Zero-Preview focuses on enhancing AI's reasoning abilities, especially excelling in mathematical logic and code writing.
🛠️ Users can experience GLM-Zero-Preview for free on the Zhiyuan Qingtalk platform, supporting text and image uploads, and outputting the complete reasoning process.
📈 As the training volume increases, the performance of GLM-Zero-Preview in deep reasoning steadily improves, showcasing the importance of reinforcement learning.
Details link: https://chatglm.cn/main/gdetail/676411c38945bbc58a905d31?lang=zh
4. Baidu Releases 2024 Annual AI Prompt Word - "Answer"
At the end of 2024, Baidu released the annual AI prompt word "Answer," reflecting people's reliance on and expectations from AI. As people frequently seek answers from AI, terms like "answer" and "why" reveal societal emotions and personal confusion. Baidu demonstrates how AI has integrated into daily life through the analysis of high-frequency prompt words, becoming a channel for thoughts and emotions.
【AiBase Summary:】
🤖 AI has become an important tool for people seeking answers in their lives, reflecting societal emotions and confusion.
🔍 High-frequency prompt words reveal common issues and desires in life, work, and emotions.
🌟 Baidu emphasizes that AI will continue to be a partner for humanity, exploring future possibilities and unknown territories together.
5. Tongyi Releases 2024 Young People's AI Usage Trends Report: Higher Attention to AI Among 85 and 90s Generations
According to the "2024 Young People's AI Usage Trends Report," AI applications have widely penetrated various aspects of life, particularly in work, learning, and creative expression. Post-95s, women, and corporate managers show the highest levels of attention towards AI. Over 80% of respondents indicated a high level of concern for AI tools, with nearly half using AI daily, indicating that AI has become an indispensable part of life.
【AiBase Summary:】
🧑🎓 Post-95s, women, and corporate managers show a significant increase in attention towards AI, with over 80% of respondents highly concerned about AI tools.
🎨 AI is widely used in creative expression and entertainment activities, with young people eager to try AI-generated content.
🔍 Despite increasing expectations for AI, concerns about data privacy are also rising, necessitating vigilance.
6. OpenAI CEO Announces New Technology Products for 2025, AGI and Adult Mode Spark Discussions
OpenAI CEO Sam Altman announced the launch of several new technology products in 2025, particularly General Artificial Intelligence (AGI) and intelligent agent functions, attracting widespread attention. The release of new products reflects OpenAI's ongoing innovation in the AI field, especially in response to user feedback, showcasing the company's sensitivity to market demands. The introduction of adult mode has sparked heated discussions online, with expectations for a more open content generation experience.
【AiBase Summary:】
🌟 OpenAI plans to launch new products such as AGI and intelligent agents in 2025, showcasing its ongoing innovation in the AI field.
💬 The adult mode has garnered attention, with expectations for a more open content generation experience.
📈 Altman's technology release is based on user feedback, reflecting OpenAI's emphasis on user needs in product development.
7. Zhiyuan Robotics Open Sources the World's First Million Real Machine Dataset - AgiBot World
Zhiyuan Robotics, in collaboration with multiple organizations, has open-sourced the AgiBot World dataset, the world's first million real machine dataset based on real-world scenarios, aimed at advancing humanoid robot technology. The scale and quality of this dataset surpass existing similar products, greatly facilitating the training and application of robotic large models.
【AiBase Summary:】
🌍 AgiBot World is the world's first million real machine dataset based on a global real scenario, supporting generalized and universal robotic large model training.
📦 The dataset covers five core scenarios, including home, dining, and industrial, containing over 3,000 real items and more than 80 skill videos.
📈 Zhiyuan Robotics plans to open-source tens of millions of simulation data in the future to promote the widespread application of humanoid robot technology.
Details link: https://github.com/OpenDriveLab/agibot-world
8. Hugging Face Launches SmolAgents: Create Intelligent Agents with Three Lines of Code, Simplifying AI Development
Hugging Face's SmolAgents toolkit brings revolutionary changes to AI development, making the creation of intelligent agents unprecedentedly simple and efficient. With just three lines of code, developers can quickly build powerful intelligent agents using pre-trained models, significantly lowering the development barrier. The lightweight design and intuitive API of SmolAgents allow developers of all skill levels to easily get started and complete tasks quickly.
【AiBase Summary:】
🚀 SmolAgents simplifies the creation of intelligent agents with three lines of code, lowering the development threshold.
📊 This toolkit utilizes pre-trained models and supports functions such as language understanding, intelligent search, and dynamic code execution.
💻 SmolAgents is suitable for various development scenarios, quickly completing tasks and is ideal for individual developers and small teams.
Details link: https://github.com/huggingface/smolagents
9. Shanghai Adds 9 Newly Registered Generative AI Services
The Shanghai Cyberspace Administration recently announced the registration of 9 new generative AI services, aimed at promoting innovation and standardized applications of generative AI in the city. This registration brings the total number of registered services to 63, emphasizing that all online services must indicate their registration numbers to enhance transparency and user trust. The newly registered services include Wuyou Smart Face and AI Synchronized Speaking Practice, aimed at providing users with a safer and more reliable service environment.
【AiBase Summary:】
📈 Shanghai adds 9 generative AI services, bringing the total registration to 63, promoting healthy industry development.
🔍 All online generative AI applications must indicate their registration numbers to enhance service transparency.
💡 New services include Wuyou Smart Face and AI Synchronized Speaking Practice, aimed at providing users with a safe and reliable experience.
10. Extremely Expensive! OpenAI's o3 Model Costs Up to $1000 Per Query!
The recently launched o3AI model by OpenAI is considered its most powerful AI product, but the operating costs are shocking, with a single task costing over $1000. The o3 scored 87.5% in the ARC-AGI benchmark test, nearly three times that of the previous generation o1 model. However, this significant performance improvement comes with enormous costs, raising concerns about its economic viability in the industry.
【AiBase Summary:】
💸 The o3AI model's cost per query exceeds $1000, indicating its high operating expenses.
📊 In the ARC-AGI benchmark test, the o3 scored 87.5%, nearly three times that of the previous generation o1 model.
🔍 Currently, the o3 has not been released to the public, and a "mini version" is expected to launch in January next year.
11. Nvidia Successfully Acquires Run:ai and Decides to Open Source Its GPU Management Software
Nvidia has recently completed the acquisition of the Israeli software company Run:ai, aiming to enhance the management efficiency of AI cloud computing. Although the specific acquisition amount has not been disclosed, the deal is valued at approximately $700 million. Nvidia announced that it will open source Run:ai's software to support a broader AI ecosystem. Run:ai's software can efficiently schedule Nvidia GPU resources, optimizing AI computing performance.
【AiBase Summary:】
🌟 Nvidia has completed the acquisition of Run:ai and announced plans to open source its software to promote AI technology development.
💻 Run:ai's software can effectively schedule Nvidia GPU resources, enhancing AI computing efficiency.
🤝 Run:ai will continue to support customers, aiming to maximize the efficiency of AI infrastructure usage.