Welcome to the 【AI Daily】 column! This is your guide to exploring the world of artificial intelligence every day. Here we present you with hot content in the AI field, focusing on developers to help you gain insight into technical trends and understand innovative AI product applications.
Fresh AI products click to learn more: https://top.aibase.com/
1. Xiaomi transfers multiple ‘Kimi’ trademarks to Moon's Dark Side
Xiaomi has transferred the ‘Kimi’ trademarks to Moon's Dark Side, a company focused on AI assistants. This move might be part of Xiaomi’s strategic adjustment to optimize resources while supporting Moon's Dark Side in expanding its intelligent assistant business.
【AiBase Summary:】
🌟 Xiaomi has transferred multiple ‘Kimi’ trademarks to Beijing Moon's Dark Side Technology Co., Ltd., optimizing non-core trademark resources.
🤖 The Kimi intelligent assistant launched by Moon's Dark Side went online in 2023; this trademark transfer supports its development.
🔍 This transaction may bring better market development opportunities for both parties, promoting the development of the intelligent assistant field.
2. Microsoft Bing launches new video creation tool Bing Video Creator, allowing users to easily generate AI videos
Microsoft Bing has launched the Bing Video Creator based on the OpenAI Sora model. Users can generate short videos for free through text prompts, but it currently only supports mobile devices and has a slower generation speed.
【AiBase Summary:】
✨ New feature: Bing Video Creator is now available for free for the first time, allowing users to generate short videos just by simple text descriptions.
📱 Limitation: Currently only supported on mobile devices, not available on desktops, affecting some user experiences.
💰 Incentive mechanism: Users can earn points through search or shopping to generate 10 videos for free before needing to pay to continue generating.
3. ElevenLabs releases new voice interaction platform Conversational AI 2.0: AI voice assistant understands you better than real people
ElevenLabs has released Conversational AI 2.0, which has made significant breakthroughs in dialogue fluency, multilingual support, and enterprise-level application capabilities, bringing new possibilities to customer service, marketing, and content creation fields.
【AiBase Summary:】
Introduced advanced turn-taking dialogue models, accurately capturing user dialogue rhythms to avoid interruptions and improve dialogue fluency.
Supports over 32 languages with seamless switching, built-in automatic language detection function, assisting global enterprises in customer service.
Integrated RAG technology, extracting information from corporate knowledge bases to ensure professional and accurate answers.
Details link: https://elevenlabs.io/blog/conversational-ai-2-0
4. Google Gemini Live function officially lands on iOS platforms, opening up a new AI recognition experience
Google’s Gemini Live function has been launched on iOS and iPadOS platforms, supporting AI recognition of scenes and screen content, and is currently free to use. This function provides convenient information acquisition experiences through cameras and screen sharing, but it is currently only available to users in the United States.
【AiBase Summary:】
✨ Gemini Live now supports iOS/iPadOS, using AI to quickly identify objects and provide information.
📱 Screen sharing function allows users to easily share screen content, enhancing interactive experiences.
🌍 The function is currently only available in the U.S., as Google promotes broader applications of AI technology.
5. Character.AI Launches New Feature AvatarFX, Allowing Users to Create Personalized Animated Videos
Character.AI has launched the AvatarFX tool, enabling users to create custom animated videos and adding new 'Scene' and 'Flow' functions, while facing abuse issues.
【AiBase Summary:】
🌟 Character.AI launches the AvatarFX tool, allowing users to create custom animated videos.
🎬 New 'Scene' and 'Flow' functions enable users to share character creations.
⚠️ Character.AI faces litigation due to abuse incidents, and the platform has security risks.
Details link: https://blog.character.ai/character-ai-unveils-new-ways-to-create/
6. OpenAI rewrites Codex CLI in Rust, bidding farewell to Node.js
OpenAI announced that it has rewritten its AI programming tool Codex CLI from Node.js to Rust language, bringing performance optimization, enhanced security, and zero dependency installation advantages.
【AiBase Summary:】
🌟 Codex CLI migrates from TypeScript and Node.js to Rust, bringing performance optimization and security improvements.
🔒 Rust achieves zero dependency installation, supports sandbox environment execution, and enhances cross-platform compatibility.
🚀 Rust language features enable Codex CLI to become a client and server for model context protocols, performing excellently.
7. NUS Launches OmniConsistency: Low-Cost Realization of Image Style Consistency, Challenging GPT-4o!
The team from the National University of Singapore (NUS) has released the OmniConsistency project, achieving a perfect combination of image style transfer and consistency at extremely low cost through unique learning frameworks and modular architectures, providing powerful tools for developers.
【AiBase Summary:】
✨ Using paired image data to learn style migration consistency, achieving impressive results with just 2,600 pairs of high-quality images and 500 hours of GPU computing power.
🔄 Supports modular architecture, compatible with existing style LoRA modules, easily integrated into various projects.
🌟 Injecting commercial-grade capabilities into the open-source ecosystem, promoting the development of AI art creation.
Details link: https://github.com/showlab/OmniConsistency
8. Hume AI Unveils EVI 3: Voice AI Understanding Your Emotions, Faster Than GPT-4o!
Hume AI has released the third-generation voice interaction model EVI3, featuring outstanding emotional understanding and personalized interaction experiences, marking a major breakthrough in emotional interaction and natural communication in voice AI.
【AiBase Summary:】
✨ EVI3 can accurately identify emotions in user speech and generate specific styles and personalities, achieving a perfect fusion of emotional intelligence and voice interaction.
🚀 Features ultra-low latency and intelligent responses, with inference latency as low as 300 milliseconds, surpassing GPT-4o in emotional expression and naturalness.
🌐 Supports multi-scenario applications, including customer service and content creation, and will expand multi-language support to cover global markets in the future.
Details link: https://demo.hume.ai/
9. Inside Scoop: Apple Has a 150 Billion Parameter AI Model Comparable to ChatGPT but Refuses to Release It
Apple plans to open its foundational model at WWDC, but its performance is limited, and there are no public release plans for its more powerful internal AI models. Leadership disputes have caused delays in multiple AI projects, and WWDC is more of a marketing showcase.
【AiBase Summary:】
🍎 Apple plans to release an AI model with approximately 3 billion parameters, which performs small and limited, mainly supporting basic functions.
🚀 Apple has larger-scale internal AI models, with a maximum of 150 billion parameters, but they are only used for internal testing and have no public release plans.
⏳ Leadership disputes at Apple are severe, causing delays in multiple AI projects, and WWDC releases mostly minor updates rather than innovative features.
10. Google Launches AI Edge Gallery App, Enabling Offline Smartphone AI Processing
Google has launched the AI Edge Gallery app, allowing users to run complex AI models offline on their phones, enhancing privacy protection and supporting multiple AI functions, but there is still room for improvement in installation and usage experience.
【AiBase Summary:】
🌟 Google launches the AI Edge Gallery app, supporting offline operation of AI models, enhancing privacy protection.
📱 The app supports downloading Hugging Face models, providing multi-turn dialogues, visual question answering, and other AI functions, all processed locally.
🔒 Local processing solves privacy issues, especially suitable for sensitive industries such as healthcare and finance.
Details link: https://github.com/google-ai-edge/gallery
11. Cerebras Opens Its Inference API Completely, Developers Get a Million Free Tokens Per Day
Cerebras Systems announced that its inference API is now completely open, eliminating waiting list restrictions, and providing a million free tokens per day, significantly improving AI inference efficiency, particularly outstanding in real-time voice and video processing areas.
【AiBase Summary:】
🚀 Inference API is now open and provides a million free tokens daily, greatly reducing developer costs.
⚡ Inference speed is 20 times faster than GPUs, especially suitable for complex inference models and code generation tasks.
🌐 Supports mainstream open-source models, seamlessly integrating with Hugging Face and Meta platforms, simplifying the developer process.
12. Nvidia and MIT Collaborate on Fast-dLLM Framework, AI Inference Speed Improves 27.6 Times
Nvidia, MIT, and Hong Kong University jointly released the Fast-dLLM framework, significantly improving the reasoning speed of diffusion models while maintaining generation quality, providing strong support for AI applications.
【AiBase Summary:】
🌟 Rapid improvement: Achieves up to 27.6 times inference speedup through block-wise approximate KV caching mechanisms.
🔍 Innovative technology: Confidence-aware parallel decoding strategy ensures generation quality, reducing dependency conflicts.
📊 Measured performance: Balanced speed and accuracy in multiple benchmark tests, promoting the widespread application of diffusion models.
Details link: https://nvlabs.github.io/Fast-dLLM/