Welcome to the "AI Daily" column! This is your guide to exploring the world of artificial intelligence every day. Each day, we present you with the latest content in the AI field, focusing on developers to help you understand technology trends and innovative AI product applications.
New AI products Click to learn more:https://app.aibase.com/zh
1. Kuaishou's Ke Ling 2.6 Fully Launched! "Audio and Visual Together", Video, Natural Voice, Matching Sound Effects, Environmental Atmosphere
Kuaishou's Ke Ling AI has launched its first "audio and visual together" model - Ke Ling 2.6. This model can generate video, natural voice, sound effects, and environmental atmosphere in a single generation process, enhancing the user's creative experience. It provides two creative paths: "text-to-audio-visual" and "image-to-audio-visual," suitable for various content formats, including monologues, voiceovers, dialogues, and music performances.
AiBase Highlights:
🎨 Ke Ling 2.6 model achieves synchronized audio-visual generation, enhancing the user's creative experience.
🖋️ Provides "text-to-audio-visual" and "image-to-audio-visual" creative paths, convenient for various content formats.
🎤 Widely applicable to monologue, voiceover, dialogue, and music performance scenarios.
2. ByteDance Shocks the Market! Seedream 4.5 Released, Image Generation Enters the "Multi-Image Consistency" Era
The Seed team under ByteDance launched Seedream 4.5, emphasizing multi-image scene consistency and aesthetic instruction following, solving pain points in image generation and improving image quality and practicality.
AiBase Highlights:
🧠 Multi-image scene consistency improved, character, clothing, lighting, and composition highly unified.
🎨 Aesthetic instruction following significantly improved, accurately hitting style keywords.
⚙️ Hard against Flux, Midjourney v6.1, a blessing for detail lovers.
3. Doubao Mobile Assistant Urgently Removes WeChat Operation Function: Responding to Account Ban and "High-Risk Permission" Controversy
Doubao Mobile Assistant issued a statement regarding user-reported issues with WeChat usage and high-risk permission disputes, admitted the problem, clarified the situation of permission acquisition, and removed the WeChat operation function while promising to protect user privacy.
AiBase Highlights:
📱 Doubao Mobile Assistant acknowledges specific issues and responds to user feedback.
🔒 Emphasizes commitment to user authorization and privacy protection.
🚫 Removes WeChat operation function and gradually lifts banned accounts.
4. Zhiyuan Research Institute Launches the World's Most Powerful Multimodal World Model Emu3.5: Predicts the "Next Second" of the Real World with One Click!
Beijing Zhiyuan Artificial Intelligence Research Institute launched Emu3.5, a multimodal large model that encodes images, text, and videos into Token sequences, achieving a deep understanding of the real world. Its core breakthrough lies in predicting the state of the "next second" of the world, evolving from a "pixel transporter" to a "world simulator," marking the transition of multimodal large models from the "generation era" to the "world model era."
AiBase Highlights:
🧠 Emu3.5 encodes images, text, and videos into Token sequences, enabling cross-modal causal relationships and physical common sense learning.
🔄 The model focuses on predicting the state of the "next second" of the world, evolving from a "pixel transporter" to a "world simulator."
🚀 This model marks the transition of multimodal large models from the "generation era" to the "world model era," with extensive application potential.
Details: https://zh.emu.world/pages/web/landingPage
5. AI Model Discovers Smart Contract Vulnerabilities, Simulated Attack Loss Reaches $4.6 Million
A new study by MATS and Anthropic reveals the dual role of advanced AI models in the field of smart contract security. These models not only can identify and exploit vulnerabilities but also help develop more powerful defense tools.
AiBase Highlights:
🔍 Advanced AI models such as Claude Opus4.5 and GPT-5 can identify and exploit smart contract vulnerabilities.
💸 Simulated attack losses reached $4.6 million, and AI models also discovered new security vulnerabilities in experiments.
🔒 AI is not only a potential risk source but can also be used to strengthen cybersecurity measures.
6. Xiaomi AI Roadmap Revealed for the First Time: Lu Weibing Confirms Betting on "AI + Physical World", Luo Fuli Leads MiMo Large Model with a Salary of Millions
Xiaomi Group President Lu Weibing revealed the group's AI strategic direction for the first time during a live broadcast, emphasizing the deep integration of large models with physical scenarios, aiming to embed AI capabilities into hardware and services. Luo Fuli has officially joined the MiMo large model team, becoming a key part of Xiaomi's AI talent strategy. Xiaomi plans to release a multimodal model with over 100 billion parameters in 2025, which will be applied simultaneously in three scenarios: smartphones, cars, and smart home devices.
AiBase Highlights:
🧠 Xiaomi Group President Lu Weibing revealed the AI strategic direction for the first time during a live broadcast, emphasizing the deep integration of large models with physical scenarios.
💼 Luo Fuli joined the MiMo large model team, becoming a key part of Xiaomi's AI talent strategy.
🚀 Xiaomi plans to launch a multimodal model with over 100 billion parameters in 2025, which will be applied simultaneously in smartphones, cars, and smart home devices.
7. Google Releases Workspace Studio: Full Launch of AI Agent Building Tool
Google released Workspace Studio, a no-code AI agent building tool, allowing enterprise employees to quickly create automated processes and improve work efficiency.
AiBase Highlights:
🤖 Create AI agents without programming, achieving automated task processing.
🌐 Supports multi-application collaboration and third-party integration, breaking information silos.
🧠 Based on Gemini3's intelligent decision-making capabilities, enhancing enterprise efficiency.
8. DeepSeek Releases Two Major New Models, Official Version V3.2 and Speciale Go Live Simultaneously
DeepSeek released two major new models, the official version V3.2 and Speciale, simultaneously, demonstrating its strong power and innovation capability in the open-source large model field.
AiBase Highlights:
🧠 DeepSeek-V3.2 first integrates "thinking process" and "tool calling" deeply, supporting two modes running in parallel.
🧠 Speciale version is an enhanced version that "turns thinking up to maximum," pushing reasoning capabilities to physical limits.
🚀 Full platform instant update, users are unaware of the change, experiencing new capabilities seamlessly.