Welcome to the 【AI Daily】 column! Here is your daily guide to exploring the world of artificial intelligence. Every day, we present you with the hottest topics in the AI field, focusing on developers, helping you gain insights into technology trends and understand innovative AI product applications.

Fresh AI Products Click to Learn More: https://top.aibase.com/

1. WeChat Integrates DeepSeek: Tencent Responds: User Personal Information and Privacy Data Will Not Be Used

Recently, WeChat has innovated its search function by beginning a gray test integration of the DeepSeek R1 model, aiming to enhance users' AI search experience. Tencent has confirmed that users participating in the test will be able to use the model for free, enjoying a richer search service. The launch of this new feature marks a significant advancement for WeChat in search technology, and Tencent has also promised not to use users' personal information, ensuring the protection of user privacy.

image.png

【AiBase Summary:】

🧠 WeChat's search function introduces the DeepSeek R1 model, enhancing AI search capabilities for a more intelligent search experience.

🔒 Tencent promises not to use personal information in AI search, ensuring user privacy and security.

🌐 Multiple Tencent products are exploring the integration of the DeepSeek model to provide more stable and comprehensive search services.

2. Baidu Search: Full Integration of DeepSeek and Wenxin Deep Search Functionality

Baidu Search recently announced full integration of the DeepSeek and Wenxin large model deep search functionalities, aiming to improve the user search experience. The deep search function of the Wenxin large model went live on February 13, featuring multi-modal input and output capabilities, providing expert-level content responses. Meanwhile, the Baidu Wenxin agent platform will also integrate DeepSeek to offer developers more convenient model calling options.

image.png

【AiBase Summary:】

🚀 The deep search functionality of the Wenxin large model went live on February 13, providing diversified search services.

💡 Developers will access DeepSeek through the Baidu Wenxin agent platform, simplifying the creation and optimization of agents.

📅 Wenxin Yiyan will be fully free from April 1, with a new version set to launch in the coming months.

3. Developers Rejoice! Byte AI Programming Tool Trae Officially Releases Win x64 Version

The Win x64 version of the Byte AI programming tool Trae has officially been released, marking another important advancement in developer tools. This update aims to provide users with a smoother operating experience and further enhance development efficiency. Trae's Builder mode simplifies task execution through dialogue with AI, integrating multi-modal features and intelligent auto-completion to greatly optimize the development process, allowing developers to focus more on creation.

image.png

【AiBase Summary:】

🚀 Trae now supports the Win x64 version, providing users with a smoother operating experience.

🤖 Builder mode allows users to converse with AI, automatically breaking down and executing tasks, enhancing work efficiency.

💡 Advanced intelligent auto-completion predicts user intent in real-time, significantly improving development efficiency.

Details link: https://www.trae.ai/

4. Musk Announces xAI Will Launch Grok 3, Calling It "The Most Powerful AI Model on Earth"

In the context of escalating global competition in artificial intelligence, billionaire Elon Musk's AI company xAI is set to release its latest Grok3 chatbot. Musk refers to Grok3 as "the smartest AI on Earth," emphasizing its ability to surpass existing market competition tools. Grok3 has the capability to reflect on its own errors and achieve logical consistency through data analysis.

image.png

【AiBase Summary:】

🌟 Musk will launch the Grok3 chatbot this Monday, calling it "the smartest AI on Earth."

💡 Grok3 has the ability to reflect on its own errors and surpass all current competitive tools in the market.

🚀 Many countries around the world are accelerating the launch of AI chatbots, leading to increased market competition.

5. UI Design Magic! Ready AI: Generate Professional-Level Web Pages with Input Prompts

Ready AI is an impressive tool that allows users to generate professional-level web interfaces in just 30 seconds using simple text commands. Its uniqueness lies in providing real-time previews and version comparison features, making the design process more efficient. Users can freely choose frameworks, color styles, and layout structures, and even upload images for inspiration.

image.png

【AiBase Summary:】

🚀 Text commands generate dual-version designs in seconds: supports A/B comparisons and historical version rollback.

🛠️ A powerful tool for front-end interface generation: requires programming tools for full functionality.

💵 Tiered pricing plan: the free version can generate 10 complete pages.

Details link: https://readdy.ai/home

6. QQ Browser Integrates DeepSeek-R1 Full Version: Supports Real-Time Internet Search for WeChat Official Accounts

On February 16, QQ Browser officially integrated the DeepSeek-R1 model full version, aiming to provide users with a smarter and more convenient search experience. The new model features deep thinking, internet search, multi-turn dialogue, and historical record retrieval, ensuring users receive comprehensive and high-quality answers. Additionally, users can easily access this feature on both mobile and desktop, improving search accuracy and efficiency.

image.png

【AiBase Summary:】

🔗 Integrating the DeepSeek-R1 model enhances the search experience, supporting multi-turn dialogue and historical record retrieval.

📱 Users can access the DeepSeek model on mobile and desktop, providing real-time internet search, including WeChat Official Account information.

📝 Provides note-taking and text extraction features, supporting various export formats to enhance information processing efficiency.

7. Light-A-Video: Video Relighting Without Training

Light-A-Video is an innovative technology aimed at solving the temporal consistency problem in video relighting. By introducing a consistent light attention module and progressive light fusion strategy, this method effectively addresses the issue of inconsistent light sources, significantly improving video quality and temporal consistency. Experimental results show that Light-A-Video not only maintains high image quality but also ensures smooth transitions in lighting across frames, providing a new direction for future research in video relighting.

image.png

【AiBase Summary:】

🌟 Light-A-Video is a training-free technology aimed at achieving temporal consistency in video relighting.

🎥 It employs a consistent light attention module and progressive light fusion strategy to resolve light source inconsistencies in video relighting.

📈 Experiments show that Light-A-Video significantly improves the temporal consistency and image quality of relighted videos.

Details link: https://bujiazi.github.io/light-a-video.github.io/

8. From Meta! Pippo: Generate High-Resolution Multi-Angle Images from a Single Character Photo

The Pippo model recently launched by Meta Reality Labs is a groundbreaking technology capable of generating high-resolution multi-angle videos from a single ordinary photo. This innovation requires no additional input parameters; users simply provide a photo, and the system automatically generates vivid stereoscopic effects. To facilitate developers, Pippo is released in a code-only version, allowing users to train the model and apply it independently.

image.png

【AiBase Summary:】

🌟 The Pippo model can generate high-resolution multi-angle videos from a single ordinary photo without requiring additional input.

💻 Code only release, with no pre-trained weights, allowing developers to train the model and apply it independently.

🔍 The team plans to introduce more features and improvements in the future to enhance user experience.

Details link: https://github.com/facebookresearch/pippo

9. Microsoft Releases OmniParser V2.0: Converts Screenshots into Structured Formats for LLM Processing

Microsoft's OmniParser V2.0 is a new parsing tool designed to convert user interface screenshots into structured data, thereby enhancing the user experience for operations based on large language models. The tool significantly improves icon recognition accuracy and processing speed through enhanced datasets and algorithms, allowing users to operate virtual machines more efficiently.

image.png

【AiBase Summary:】

🔍 OmniParser V2.0 can convert UI screenshots into structured information, enhancing user operation experience.

⚡ The new version reduces average latency to 0.6 seconds/frame with an accuracy rate of 39.6%.

🔐 Users should be mindful of the security of input content, and developers should adhere to safety standards and ethical guidelines.

Details link: https://huggingface.co/microsoft/OmniParser-v2.0

10. The Dark Side of the Moon Decoded: Long-CoT is Key, Model Thinking Needs to Be Prolonged

Flood Sung, a researcher on the Dark Side of the Moon, delves into the development ideas of the k1.5 model and the technical insights from the OpenAI o1 model in a lengthy article. The article emphasizes the importance of Long-CoT (long chain thinking), pointing out its significant effects in training small models. Although the focus was previously on optimizing Long Context due to cost considerations, the release of OpenAI o1 prompted the team to re-evaluate their technical direction, deciding to fully promote Long-CoT research to achieve thinking capabilities closer to that of humans.

image.png

【AiBase Summary:】

🌟 Long-CoT has been proven to have significant effects in multi-digit operations training for small models, emphasizing its importance at the output end.

💡 The release of OpenAI o1 has prompted the Dark Side of the Moon to reassess technical priorities, believing that performance breakthroughs are the primary goal.

🔍 The Dark Side of the Moon has begun systematic benchmarking against the o1 model, committed to conducting substantial research in relevant fields.

Details link: https://mp.weixin.qq.com/s/sJmT-tM3A-mglZ1d4OI80A

11. 80% Accuracy! Meta Develops Non-Invasive Brain-Computer Interface That Allows Typing by Thought Alone

Meta has recently developed a non-invasive brain-computer interface device that can achieve text input by reading neural signals from the human brain. This technology utilizes a magnetoencephalography (MEG) scanner and deep learning AI model to successfully decode brain signals while typing, reconstructing complete sentences. Although the device weighs nearly half a ton, costs $2 million, and requires use in a specialized environment, its current accuracy rate has reached 80%.

image.png

【AiBase Summary:】

🧠 Meta's non-invasive brain-computer interface device allows text input through brain signals.

💰 The device weighs half a ton, costs $2 million, and requires use in a specialized environment.

📊 Currently, its accuracy rate is 80%, but improvements are needed, as it is still some distance from practical application.