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 technical trends and innovative AI product applications.
New AI products click to learn more:https://app.aibase.com/zh
1. Kuaishou's 72B Code Model Open-Sourced and Tops SWE-Bench: KAT-Dev Surpasses SWE-Bench, Marking a Breakthrough in Domestic AI Programming
The KAT-Dev-72B-Exp model developed by Kwaipilot, a subsidiary of Kuaishou, achieved a 74.6% accuracy rate in the SWE-Bench Verified benchmark test, becoming a top performer among open-source code models. The model demonstrates excellent performance in code completion, bug fixing, and system-level reengineering through an innovative training architecture and reinforcement learning mechanism, marking a milestone in domestic AI programming assistants.
AiBase Summary:
🔥 The KAT-Dev-72B-Exp model achieved a 74.6% accuracy rate in the SWE-Bench test, topping the open-source code model rankings.
💡 The model uses shared prefix trajectory and entropy shaping strategies to enhance training stability and generalization ability.
🚀 Kuaishou open-sources KAT-Dev-72B-Exp, providing global developers with free and efficient tools to promote the development of AI programming assistants.
Details: https://huggingface.co/Kwaipilot/KAT-Dev-72B-Exp
2. Hangzhou Yunchuchu Launches the World's First IP66 Protected Humanoid Robot DR02: Works in Rain, Operates in Temperatures from -20°C to 55°C
The world's first IP66-rated humanoid robot, DR02, launched by Hangzhou Yunchuchu Technology, has all-weather operation capabilities and can run stably in harsh environments, marking a key step toward the practical application of industry technology.
AiBase Summary:
✅ The world's first IP66-rated humanoid robot DR02 officially launched, featuring dust and water resistance, capable of stable operation in heavy rain.
🌡️ DR02 can operate in a wide temperature range from -20°C to 55°C, adapting to complex environments.
🔧 Modular quick-disassembly design improves maintenance efficiency, allowing rapid replacement of core components.
3. Google Chrome to Introduce Gemini New Features, Making Web Browsing Smarter
Google Chrome is about to introduce new Gemini features, making web browsing smarter, while also raising concerns about privacy protection.
AiBase Summary:
🤖 Google is developing the "Context Task" feature to enhance the chat interaction experience of Gemini in the Chrome browser.
🛒 Gemini will be able to automatically complete tedious tasks such as online grocery ordering, shortening operation time.
🔄 Google is also developing the "Multi-Instance Gemini" feature, allowing users to call AI assistants simultaneously across multiple tabs.
4. Scholars Sue Apple for Using Pirated Books to Train AI, Reigniting Copyright Controversy
Two professors in the United States have filed a lawsuit against Apple for using their works without permission to train AI models, drawing widespread attention to copyright issues in AI training. The case involves the Books3 dataset, which contains a large amount of unauthorized book texts.
AiBase Summary:
📚 Professors accuse Apple of using their books without permission to train AI models.
⚖️ If found guilty of "willful infringement," Apple could face significant compensation risks.
🌐 This case reflects the widespread controversy over the legality of copyright in AI training.
5. Liquid AI Launches LFM2-8B-A1B: Only 1.5B Parameters Activated, Achieving 4B-Level AI Speed on Mobile Devices!
Liquid AI's LFM2-8B-A1B model achieves efficient edge computing capabilities through sparse activation mechanisms, achieving 4B-level performance with only 1.5B parameters activated. The model performs well in multilingual processing, mathematical abilities, and coding tasks, and supports various deployment methods, significantly improving the feasibility of device-side AI.
AiBase Summary:
🧠 Sparse activation mechanisms reduce computational load and improve edge device performance.
📊 Pre-training data volume reaches 12T, supporting multilingual and code processing.
🚀 Supports multiple frameworks and quantized versions, facilitating rapid integration by developers.
Details: https://huggingface.co/LiquidAI/LFM2-8B-A1B
6. Apple Plans to Acquire Computer Vision Startup Prompt AI, Further Expanding into the Smart Home Field
Apple is approaching a major deal, planning to acquire the team and technology of computer vision startup Prompt AI to quickly introduce key AI capabilities. Prompt AI's Seemour product can identify people, pets, or objects through home cameras and describe them in natural language, showing great potential for application. This acquisition will help Apple strengthen its position in the smart home and visual AI technology fields, further driving innovation in the HomeKit, camera, and home security ecosystems.
AiBase Summary:
🍎 Apple is close to acquiring computer vision startup Prompt AI, enhancing its AI capabilities.
👥 Prompt AI's Seemour product can identify and describe people and objects in home environments through cameras.
🏡 This acquisition aims to strengthen Apple's presence in the smart home and visual AI technology fields.
7. AI Girlfriend App Security Collapse: Over 4 Million Users' 43 Million Private Conversations Leaked
An AI companion app suffered a security breach, leading to a large-scale leakage of user data, raising concerns about the safety and ethics of AI applications.
AiBase Summary:
🔐 Data Breach: Over 4 million users' data, 43 million messages, and 600,000 images and videos exposed.
🌐 Security Vulnerability: Unprotected Kafka Broker instances left the data vulnerable.
💰 Financial Risk: Authentication tokens may be used by hackers to hijack accounts and steal virtual currency.
8. AI Scientist DeepScientist from West Lake University Boosts Research Efficiency Dramatically
West Lake University's AI scientist DeepScientist has demonstrated its powerful capabilities in research, completing a large amount of work in a short time and achieving breakthrough results in multiple cutting-edge tasks.
AiBase Summary:
🚀 DeepScientist completed the research of three years in two weeks, demonstrating strong research capabilities.
💡 The system generates and verifies scientific ideas autonomously through a closed-loop, iterative process, forming a complete research loop.
🧠 DeepScientist successfully broke human research records in multiple cutting-edge tasks, showcasing the huge potential of AI in the field of research.
Details: https://github.com/ResearAI/DeepScientist
