Welcome to the AI Daily section! Here is your daily guide to exploring the world of artificial intelligence. Every day, we bring you the hottest topics in the AI field, focusing on developers, helping you understand technological trends, and discover innovative AI product applications.
Discover New AI Products: https://top.aibase.com/
1. Bing Introduces Generative AI Search Feature
Bing has recently launched a new search experience, placing AI-generated answers prominently on the search results page, altering the traditional arrangement of search results. This move aims to provide more comprehensive answers by combining large and small language models to more effectively meet user needs. However, this change has also sparked some user concerns.
AiBase Highlights:
🔍 AI-generated answers are placed at the most prominent position on the search results page.
🤖 The new search experience combines large and small language models.
⚖️ Balancing AI-generated content with traditional search results is necessary to ensure information accuracy and diversity.
2. Stunning Arrival! Open-Sora Plan v1.2 Released, Clarity and Inference Speed Soared
The Open-Sora Plan v1.2 version introduces a new 3D full attention architecture, enhancing the understanding of the physical world. The update brings a new 3D full attention architecture, upgrades text-to-video capabilities, improves clarity and consistency, perfectly integrates space and time, and significantly boosts inference speed. The Open-Sora team has open-sourced code, data, and models, committed to making everyone a god of video creation.
AiBase Highlights:
🌟 The new 3D full attention architecture allows AI to have a qualitative leap in understanding the physical world, 360-degree understanding of the three-dimensional world.
🎥 Text-to-video capabilities are upgraded, typing text can present vivid video scenes.
⏱️ Perfect integration of space and time, significant improvement in spatial representation and temporal fluency of videos.
Details Link: https://top.aibase.com/tool/open-sora-plan-v1-2
3. Divine Showdown! Mistral Large2 Suddenly Open-Sourced: 123 Billion Parameters, Comparable to Llama3.1
Mistral AI has launched its flagship model Mistral Large2, with 123 billion parameters, a huge 128k context window, and outstanding performance and cost efficiency. Users can access the new model through La Plateforme, widely applied on cloud service platforms.
AiBase Highlights:
🌟 Mistral Large2 has a 128k context window, supports up to ten languages and over 80 programming languages.
📈 Achieved an accuracy of 84.0% in the MMLU benchmark test, with outstanding performance and cost efficiency.
💻 Users can access the new model through La Plateforme, widely applied on cloud service platforms.
Details Link: https://console.mistral.ai/
4. Tencent Zhiying PC Client Launches "Smart Canvas" Feature
Tencent Zhiying PC client recently introduced a new feature—"Smart Canvas," providing users with a variety of practical image editing functions, combined with AI painting technology, making drawing easier. This feature is especially suitable for users who need to re-create, cut out, erase, and expand AI painting images. Users can now log in to the Zhiying homepage to experience these new features.
AiBase Highlights:
🎨 Smart Canvas combines AI painting technology, providing a variety of practical image editing functions, allowing users to easily re-create, cut out, erase, and expand images.
🖌️ Users can choose canvas sizes and upload images, use rich material stickers and tools for editing, and also conduct AI creation.
🔍 Smart Canvas provides image AI adjustment functions, including cropping, cutting out, erasing, expanding, local redrawing, and lossless high-definition, meeting various creative and professional needs.
5. Kingsoft WPS AI Launches "AI Companion Writing" Feature
Kingsoft recently introduced the AI Companion Writing feature in WPS AI, aiming to enhance users' writing efficiency and quality. Users can enable this feature through the WPS Office interface, enjoy smart suggestions and continuation services, and easily express inspiration. AI Companion Writing also provides diversified content generation and Chinese poetry citation support, enhancing writing coherence. The upgrade of WPS AI2.0 further promotes the application of artificial intelligence in the office field.
AiBase Highlights:
✨ Enhance writing efficiency and quality, smart assist users in writing.
📚 Various scene roles meet the writing needs of different users.
💡 Provide smart suggestions, continuation services, and diversified content generation, support Chinese poetry citation.
6. Stable Video 4D Emerges, One Click to Turn Your Video into a Panoramic Blockbuster!
Stable Video4D is a groundbreaking video processing tool launched by Stability AI, which can turn ordinary videos into all-round panoramic blockbusters. It quickly generates multi-angle videos while maintaining picture consistency, which will impact game development, video editing, and VR production fields. The future may change the way of watching movies, bringing a new interactive experience.
AiBase Highlights:
🎥 Stable Video4D can turn ordinary videos into panoramic blockbusters, showing multi-angle details.
🔮 Quickly generate multi-angle videos, maintain picture consistency, and have broad application prospects.
🌌 The future may change the way of watching movies, bringing a new interactive experience.
Details Link: https://huggingface.co/stabilityai/sv4d
7. AI Music Generation Tool Udio Updates V1.5 Model, Significant Improvement in Sound Quality
Last night, the AI music generation tool Udio brought a series of impressive updates, with the V1.5 model's sound quality significantly improved, providing music creators with a clearer and richer auditory experience. New features include key tone control, multi-language support, and more, broadening the user base. Product enhancements include a dedicated creation page, downloading music clips, providing a more personalized and efficient creation environment.
AiBase Highlights:
✨ V1.5 model sound quality significantly improved, providing a clearer and richer auditory experience.
🎵 New features include key tone control and multi-language support, meeting the needs of creators.
🔧 Product enhancements include a dedicated creation page, downloading music clips, providing a more personalized and efficient creation environment.
Details Link: https://top.aibase.com/tool/udio
8. Comparable to GPT-4o! Fudan University Launches SpeechGPT2, Capable of Understanding Your Emotions
SpeechGPT2 is an innovative large language model proposed by the research team of Fudan University, with cross-modal speech understanding and generation capabilities. Although it demonstrates strong task execution capabilities, it still faces challenges in noise robustness and sound quality stability. The team plans to open-source technical reports, code, and model weights in the future to promote further development and improvement of the technology.
AiBase Highlights:
🔑 SpeechGPT2 is a new large language model with cross-modal speech understanding and generation capabilities.
🔑 SpeechGPT2 undergoes a three-stage training strategy, including modality adaptation pre-training, cross-modal instruction fine-tuning, and modality chain instruction fine-tuning.
🔑 SpeechGPT2 shows strong capabilities, performing well on text tasks, cross-modal tasks, and spoken dialogue tasks.
Details Link: https://top.aibase.com/tool/speechgpt2
9. Reddit Launches "Paywall", Blocking Search Engines and AI Bots from Freely Crawling Content
Reddit has recently taken a noticeable action by restricting major search engines and AI bots from accessing its content, requiring payment to obtain it. This move has resulted in search engines other than Google being unable to easily access the latest Reddit content, sparking widespread attention and discussion.
AiBase Highlights:
🌐 Paywall Launched: Reddit restricts search engines and AI bots from accessing content, requiring payment to obtain.
🤖 Google独占资源: Only Google can access the latest results through "site:reddit.com", other search engines are excluded.
💰 Data Monetization Strategy: Reddit strengthens data protection, raises API fees, seeks new revenue sources to attract investors.
10. Nvidia AI Launches ChatQA2, Long Text Understanding and RAG Capabilities Comparable to GPT-4
In the rapid development of artificial intelligence, the ability to understand long text context and retrieve augmented generation (RAG) has become crucial. Nvidia AI's latest research—ChatQA2 model, is born to meet this challenge. ChatQA2 achieves comparable long text understanding and RAG performance to GPT-4-Turbo by expanding the context window and implementing a three-stage instruction adjustment process.
AiBase Highlights:
⚙️ ChatQA2 significantly improves instruction following capabilities and long text understanding by expanding the context window to 128K tokens.
🔍 ChatQA2 surpasses GPT-4-Turbo in the InfiniteBench evaluation, showing comprehensive capabilities in multiple tasks.
💡 ChatQA2 solves key issues in the RAG process, improving retrieval accuracy and efficiency.
Details Link: https://arxiv.org/abs/2407.14482
11. Baichuan Intelligence Completes A-Round Financing of 5 Billion Yuan, Valuation Reaches 20 Billion Yuan
Baichuan Intelligence recently completed an A-round financing, with a total financing amount of 5 billion yuan, and the valuation soared to 20 billion yuan. This marks an important capital support for large model startups, showing the vitality and potential of the industry development.
AiBase Highlights:
🚀 Large model startup Baichuan Intelligence completes 5 billion yuan A-round financing, valuation reaches 20 billion yuan, attracting state-owned background industrial investment funds.
💡 Baichuan Intelligence stands out in the medical AI field, Baichuan3 model surpasses GPT-4, proposes AI medical L0-L5 hierarchical development route.
💰 Changes in the financing landscape of the large model industry, state-owned background funds become an important source of funds, the company adopts a super model + super application dual-wheel drive strategy.
12. Nvidia Launches Minitron Small Language Model
Nvidia's latest Minitron small language model has caused a sensation in the artificial intelligence field. This series of models has increased training speed by 40 times, significantly reduced training costs through pruning and knowledge distillation techniques, and has been open-sourced on Huggingface, promoting the popularization of AI technology.
