Welcome to the 【AI Daily】column! This is your daily guide to exploring the world of artificial intelligence. Every day, we present you with the hottest AI news, focusing on developers and helping you understand technology trends and innovative AI product applications.
Discover new AI products Learn More: https://top.aibase.com/
1、DeepSeek Releases Parallel Strategy Upgrade on its Fourth Day of Open Source: DualPipe and EPLB Technologies Revolutionize Large Model Training
On the fourth day of its open-source initiative, DeepSeek introduced optimized parallel strategies, focusing on the DualPipe bidirectional pipeline parallel algorithm and the dynamic load balancer EPLB. These technologies aim to address core challenges in large-scale language model training, significantly improving computational efficiency and resource utilization.
【AiBase Summary:】
🔄 The DualPipe algorithm implements a bidirectional data flow pipeline, increasing computational throughput and suitable for model training with hundreds of billions to trillions of parameters.
⚖️ The EPLB dynamic load balancer solves the hot expert problem in mixed-expert models, increasing overall utilization to over 92%.
📊 The compute-communication overlap optimization tool builds a spatiotemporal efficiency model, reducing end-to-end training time by approximately 15%.
Details: https://github.com/deepseek-ai/DualPipe
2、Alibaba Launches 2026 Spring Recruitment, Opening 3000 Positions with Nearly 50% Related to AI
Alibaba has officially launched its 2026 spring internship recruitment, opening over 3000 positions, with nearly 50% related to artificial intelligence. The proportion of AI positions is even higher in some departments, reaching 65% at AutoNavi and over 80% at Alibaba Cloud. This spring recruitment spans multiple departments, demonstrating Alibaba's emphasis on AI technology and its continued focus on technical roles, especially in the AI field.
【AiBase Summary:】
🤖 Alibaba launches 2026 spring internship recruitment, opening 3000 positions, with nearly 50% related to AI.
📈 In departments like AutoNavi and Alibaba Cloud, the proportion of AI positions is significantly higher, reaching 65% and 80%, respectively.
💼 Alibaba's AI To C business has begun large-scale recruitment, with 90% of positions concentrated in AI technology and product development.
3、ElevenLabs Releases Scribe Speech-to-Text Model, Achieving Record Accuracy with 96.7% in English
ElevenLabs recently launched its latest speech-to-text model, Scribe v1, claiming the highest accuracy across multiple languages. Supporting 99 languages, the model can accurately distinguish up to 32 different speakers in complex audio environments. Scribe is priced at $0.40 per hour and offers a 50% discount for the next six weeks.
【AiBase Summary:】
🌟 Scribe v1 is ElevenLabs' latest speech-to-text model, achieving record-high accuracy across multiple languages.
🗣️ Supports 99 languages and can distinguish up to 32 different speakers, adapting to complex audio environments.
💰 Currently priced at $0.40 per hour, with a 50% discount for the next six weeks; a low-latency version is under development.
Details: https://elevenlabs.io/blog/meet-scribe
4、Microsoft Releases Phi-4 Multimodal and Mini Models, Upgrading Speech, Vision, and Text Processing
Microsoft recently introduced new models in the Phi-4 series, including Phi-4 multimodal and Phi-4 mini, significantly enhancing the processing capabilities of AI applications. The Phi-4 multimodal model integrates speech, vision, and text processing, with 56 million parameters, and excels in various benchmark tests, particularly in automatic speech recognition and translation tasks. The Phi-4 mini focuses on text processing, with 38 million parameters, and also demonstrates excellent performance.
【AiBase Summary:】
🎤 The Phi-4 multimodal model is Microsoft's first unified architecture model integrating speech, vision, and text processing, with 56 million parameters, outperforming many competitors.
📊 The Phi-4 multimodal model excels in visual processing and mathematical reasoning, effectively understanding documents and charts and performing optical character recognition.
📝 The Phi-4 mini model focuses on text processing, with 38 million parameters, excelling in tasks such as text reasoning and programming, surpassing several popular large language models.
5、Hugging Face Launches FastRTC: Making Real-Time Speech and Video Application Development a Breeze
Hugging Face recently launched FastRTC, an open-source Python library designed to simplify the development of real-time audio and video AI applications. By automating complex real-time communication functions, the library allows developers to create basic real-time applications in just a few lines of code, significantly reducing development time.
【AiBase Summary:】
🎉 Hugging Face launches FastRTC, an open-source Python library designed to simplify the development of real-time audio and video AI applications.
⚡ FastRTC can accomplish work that previously took weeks in just a few lines of code, enabling existing Python developers to easily build speech and video functionality.
🌟 The release of this library presents significant opportunities for the AI community, fostering more natural human-computer interaction and helping businesses meet user needs more quickly.
Details: https://huggingface.co/fastrtc
6、FLORA Node-Based AI Canvas: Simplifying the Creative Workflow from Story Analysis to Visual Content Generation
FLORA's recently launched node-based AI canvas is a tool designed for creative professionals, aiming to streamline the creative process by integrating multiple AI functions. Its core is a node-based system where users can create independent nodes to handle different tasks. FLORA's story analysis and prompt generation, character design tools, and team collaboration features make creative work more efficient and flexible.
【AiBase Summary:】
🖌️ The node-based system allows users to independently handle different creative tasks, improving efficiency.
📖 Story analysis and character design tools generate detailed prompts for use with advanced AI image generators.
🤝 Supports real-time team collaboration, with a user-friendly interface suitable for users with limited technical backgrounds.
7、Imminent Release? OpenAI GPT-4.5 Appears in Android App Beta
OpenAI is preparing a preview version of its next-generation language model, GPT-4.5, generating considerable attention. The model will be launched as an experimental option in the ChatGPT Android app, initially available exclusively to Pro subscribers. While specific features remain unclear, GPT-4.5 is expected to succeed the free version of ChatGPT, potentially with higher usage limits.
【AiBase Summary:】
🚀 GPT-4.5 is about to be released, initially targeting Pro subscribers.
🔍 The model appears as an experimental option in the ChatGPT Android app, with specific features yet to be revealed.
💰 The Pro subscription costs $200 per month, offering more features and fewer restrictions.
8、ByteDance's AI Smart Assistant Doubao App Launches "Bring Photos to Life" Feature
ByteDance's Doubao app has launched a "Bring Photos to Life" feature, designed to transform static old photos into dynamic videos. Users simply upload a photo and describe the action to easily achieve this transformation. This feature not only adds vivid color to users' memories but also gives new life to precious moments, reflecting a combination of technology and emotion.
【AiBase Summary:】
📸 This feature allows users to transform static old photos into vivid dynamic videos, meeting users' needs for dynamicizing old photos.
💡 Easy to operate, users only need to upload a photo and describe the action to generate dynamic effects.
❤️ Doubao app hopes to use this feature to help users have a cross-time and space dialogue with their past selves, preserving beautiful moments.
9、Bilibili's Text-to-Speech Model IndexTTS: Supports Pinyin Correction for Chinese Pronunciation and Precise Pause Control
Bilibili's IndexTTS model is a GPT-style text-to-speech system based on XTTS and Tortoise, featuring unique pinyin correction for Chinese pronunciation and precise pause control. Trained on tens of thousands of hours of data, IndexTTS excels in word error rate and audio quality evaluations, surpassing many popular TTS systems and demonstrating industry-leading performance. 
【AiBase Summary:】
🌟 IndexTTS is a GPT-style TTS model based on XTTS and Tortoise, capable of correcting Chinese pronunciation and controlling pauses.
📊 The system has been trained on tens of thousands of hours of data, surpassing many existing popular TTS systems, demonstrating industry-leading performance.
🔍 IndexTTS excels in various evaluations, with word error rates and audio quality superior to other models, showcasing its strong advantages.
Details: https://github.com/index-tts/index-tts
10、Kuaishou's Keling AI Sees 113% Month-over-Month Growth in Global Monthly Active Users in January
According to the latest data, Kuaishou's Keling AI saw a 113% month-over-month increase in global monthly active users in January. UBS points out that online entertainment and education are core areas for AI applications, and Kuaishou, with its self-developed Keling AI, is a leader in global video generation models. Recently, Kuaishou launched a multi-image reference function, allowing users to upload multiple reference images, further enhancing user experience and creative freedom.
【AiBase Summary:】
📊 Kuaishou's Keling AI saw a 113% month-over-month increase in global monthly active users in January.
🎓 Online entertainment and education are key scenarios for AI implementation.
🖼️ Keling AI's newly launched multi-image reference function supports users uploading multiple reference images.
11、University Professor Says AI Essays Will Receive a Zero Score
With the rapid development of artificial intelligence technology, AI tools have become assistants for college students in completing reports and papers. However, some university professors point out that academic misconduct exists among students who rely on AI-generated content, leading schools to introduce policies that essays submitted by students using AI will receive a zero score. This measure aims to emphasize academic integrity and discourage over-reliance on technology.
【AiBase Summary:】
📚 Some university professors point out that academic misconduct exists among students who rely on AI-generated content.
🚫 Schools have introduced policies that essays submitted by students using AI will receive a zero score.
🧠 Netizens have mixed reactions, with support and concerns coexisting, emphasizing the importance of academic integrity.
12、19-Year-Old Female Go Player Heavily Penalized and Banned for 8 Years for AI Cheating
The Chinese Go Association has severely punished professional Go player Qin Siyue for cheating in the National Go Championship, revoking her professional rank and banning her for eight years. Qin Siyue carried a mobile phone during the competition, using an AI program to cheat, a serious offense, and concealed the facts when questioned. This punishment aims to maintain fairness and justice in the Go industry, warning players to abide by competition rules and prevent cheating.
【AiBase Summary:】
📱 Qin Siyue used a mobile phone and an AI program to cheat during the competition, a serious offense.
🚫 The Chinese Go Association decided to revoke Qin Siyue's professional rank and cancel her competition results.
⏳ Qin Siyue is banned from participating in Go events and activities for eight years to maintain fairness in the industry.
13、Anthropic Opens Claude AI GitHub Integration, Boosting Developer Code Efficiency
Recently, Claude's head of relationships, Alex Albert, announced the full opening of Claude's GitHub integration to all users, including free users, Pro users, and team users. The launch of this new feature means that developers will have stronger tool support in their daily coding, testing, and debugging work, enabling more efficient project development.
【AiBase Summary:】
🚀 Claude AI now offers GitHub integration, available to all users, boosting development efficiency.
💻 Developers can synchronize code repositories to Claude, enjoying stronger code analysis and debugging support.
⚠️ Free users should be mindful of quota consumption, while Pro users have better control over usage.
