Welcome to the "AI Daily" section! 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.
Fresh AI products Click to learn more:https://app.aibase.com/zh
1、The AI Personality Test Product SBTI Goes Viral Online: Focus on Abstract Tags and AI Synthesis Technology
The AI personality test product SBTI quickly went viral online due to its absurd "abstract tags" and deconstructive expression. Its core lies in using AI synthesis technology for content generation, rather than a rigorous psychological scale. Although lacking academic support, its lightweight interaction design and entertainment exploration have successfully attracted a large number of users, while also sparking discussions about the ethics and rigor of AI evaluation tools.
【AiBase Highlights:】
🧠 The SBTI app generates abstract tags through AI synthesis technology, challenging traditional personality testing tools like MBTI.
🎭 The product attracts internet subculture groups with self-deprecating tags like "Beauties" and "Mao Le," achieving entertainment exploration.
⚠️ Developers emphasize that SBTI is not suitable for scientific assessment, sparking discussions on the ethical issues of non-professional background AI evaluation tools.
2、Alibaba's Black Tech Shakes the Market! 0.6B Small Model "Modified" into 17B MoE, Only 5% Parameters Activated, Runs 30token/s on CPU!
The Alibaba International Digital Commerce team launched the Marco-Mini-Instruct model, which upgrades a small model into a MoE architecture through Upcycling technology, activating only 0.86B parameters, with high reasoning efficiency, even running on ordinary CPUs, greatly lowering the local deployment threshold, and surpassing 4B-level Dense models in performance, demonstrating the potential of the MoE architecture.
【AiBase Highlights:】
🧠 Upcycling technology upgraded the Qwen3-0.6B-Base model into a MoE architecture, achieving efficient training.
⚡ Marco-Mini-Instruct can achieve a reasoning speed of 30token/s on CPU, lowering the deployment threshold.
🚀 Activating only 0.86B parameters yet surpassing 4B-level Dense models, verifying the potential of the MoE architecture.
3、AI Music Enters the "Cover" Era! MiniMax Launches Music 2.6: Introduces New Cover Function and Agent Skills
MiniMax's Music 2.6 model made significant optimizations at the technical level, solving pain points in the AI music generation process, and introduced new interactive features such as "Cover" creation and AI Agent skills, further enhancing the controllability and practicality of AI in music creation.
【AiBase Highlights:】
🧠 Core upgrade: smarter, smoother, better-sounding
🎧 Innovative features: launch of "Cover" and AI Agent skills
🔄 Creators' benefits: global free beta testing starts
4、Kouzi 2.5 Major Upgrade! Equipping AI Agents with Cloud Computers + Cloud Phones + Independent Email, Also Building a "Parallel World"
Kouzi 2.5 version, through the launch of the Agent World platform, transforms AI agents from execution tools into independent digital partners, providing infrastructure such as cloud computers, cloud phones, and independent emails, while building an open agent collaboration and evolution platform, driving the industry transformation of AI from auxiliary execution to autonomous partners.
【AiBase Highlights:】
🧠 The Agent World platform provides AI agents with independent identity, long-term memory, and a dedicated virtual world, allowing them to survive, learn, and collaborate autonomously.
💻 The introduction of cloud computers and cloud phones allows AI agents to run code scripts, browse web pages, process files, and support downloading and installing various native apps.
📧 Agents obtain an independent email identity (@coze.email), enabling communication and collaboration with other agents or external systems, expanding application boundaries.
5、Core Talent Loss in ByteDance Seed Team: 70 People Left in One Year, Tencent and Alibaba Are Main Destinations
The article points out that the AI core department Seed team under ByteDance is facing severe talent loss, with nearly 70 technical talents leaving in the past year, mainly moving to big companies such as Tencent and Alibaba. At the same time, the talent overflow has triggered an "ByteDance-style" entrepreneurship trend, with over 30 AI companies founded by former ByteDance employees having received funding. Although ByteDance has introduced high-value stock options to retain talent, it has still been unable to stop the active migration of talent in the industry's hot spots.
【AiBase Highlights:】
🧠 In one year, nearly 70 technical talents left the ByteDance Seed team, mainly moving to Tencent and Alibaba.
💼 The talent overflow has triggered an "ByteDance-style" entrepreneurship trend, with over 30 AI companies receiving funding.
💰 ByteDance introduced high-value stock options to retain core talents, but failed to stop the talent flow.
6、Google Gemini Launches Interactive Simulation Function: Make Complex Concepts "Move"
Google Gemini has launched a new feature that can generate interactive 3D models and physical simulation scenarios, transforming AI from a content generator into an intuitive teaching tool. This feature enhances users' understanding of physical processes and three-dimensional space, and holds significant importance in education, engineering, and research fields.
【AiBase Highlights:】
✨ The new feature allows users to understand complex concepts through interactive 3D models, improving learning efficiency.
🔄 Users can adjust parameters in real-time and observe changes in physical simulations, enhancing the intuitive experience.
🚀 Google Gemini further solidifies its leading position in the multi-modal interaction field.
7、DeepSeek V4 Will Be Released in Late April: Trillion-Parameter Scale and Deep Adaptation to Domestic Chips
The release of DeepSeek V4 marks a key breakthrough for domestic large models in the trillion-parameter track, and achieves deep adaptation to domestic chips, providing important support for building an independent computing foundation.
【AiBase Highlights:】
🧠 DeepSeek V4 will achieve a trillion-parameter scale and a million-context window upgrade.
⚙️ For the first time, it achieves deep adaptation to domestic chips such as Huawei Ascend, marking the Chinese AI industry's departure from CUDA ecosystem dependence.
💰 Domestic tech giants have already pre-ordered large quantities of the next-generation AI chips, pushing chip prices up by about 20%.
8、Liang Wenfeng Reveals DeepSeek V4 Will Be Released in Late April
DeepSeek founder Liang Wenfeng announced that the new flagship large model DeepSeek V4 will be officially released in late April 2026. Before the official release, DeepSeek has introduced a layered mode design, including fast mode and expert mode, targeting daily conversations and complex logic processing. However, the platform recently experienced large-scale service abnormalities, which are seen as the "pain period" during the transition between old and new models. At the same time, Yao Shunyu's team under Tencent will also release the Huan Yuan large model, and two major domestic AI giants will compete in base model capabilities.
【AiBase Highlights:】
🔥 DeepSeek V4 will be released in late April 2026.
🧠 Introduced a layered mode design, including fast mode and expert mode.
⚠️ The platform recently experienced large-scale service abnormalities, seen as the "pain period" during the transition between old and new models.
⚔️ Tencent will release the Huan Yuan large model, competing with DeepSeek V4.
