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 applications.
Discover new AI products Learn More: https://top.aibase.com/
1. Manus, the World's First General-Purpose AI Agent, Takes Off; Invitation Codes Resell for Up to 50,000 Yuan
Manus, the world's first general-purpose AI agent, has garnered significant attention in the tech world. Manus possesses the ability to think independently and execute complex tasks, delivering complete results and demonstrating powerful versatility. It can handle daily tasks, as well as conduct in-depth market research, personalize travel plans, and much more. In the secondary market, Manus invitation codes are priced anywhere from 999 yuan to 50,000 yuan, with some sellers refusing to negotiate, highlighting the product's scarcity.
【AiBase Summary:】
🚀 Manus possesses the ability to think independently and execute complex tasks, demonstrating powerful versatility and execution capabilities.
📊 Manus' applications in various fields, such as market research and travel planning, showcase its practicality and efficiency.
🏆 Manus set a new record in the GAIA benchmark test, significantly outperforming similar products and proving its leading position.
Details: https://manus.im/
2. Alibaba Open-Sources QwQ-32B Inference Large Language Model, Matching DeepSeek-R1 Performance with Lower VRAM Requirements
Alibaba's Qwen team has released the open-source large language model QwQ-32B, aiming to enhance its performance on complex problem-solving tasks through reinforcement learning. Based on 32 billion parameters and an extended context length of 131,072 tokens, this model can match the performance of larger models in benchmarks like mathematics and programming, while requiring less VRAM.
【AiBase Summary:】
🚀 QwQ-32B utilizes reinforcement learning to improve its ability to solve complex problems.
💡 It performs comparably to larger models in mathematics and programming benchmarks, while requiring less VRAM.
🧠 It features extended context length and agentic capabilities, and future research will continue to explore the potential of reinforcement learning.
Details: https://qwenlm.github.io/blog/qwq-32b/
3. OpenAI Announces Phased Rollout of GPT-4.5 to All ChatGPT Plus Users
OpenAI recently announced a phased rollout of its latest AI model, GPT-4.5, to ChatGPT Plus users. While the model shows significant improvements in conversational abilities, it still falls short in complex reasoning. The high cost of $150 per million tokens raises concerns about its widespread adoption.
【AiBase Summary:】
💬 GPT-4.5, OpenAI's latest and largest AI model, will be gradually rolled out to ChatGPT Plus users.
⚖️ While GPT-4.5 shows significant improvements in conversational abilities, it still has shortcomings in complex reasoning.
💰 The cost of using GPT-4.5 is a high $150 per million tokens, raising concerns about its widespread adoption.
4. Doubao Launches Deep Reasoning Mode: Visualizing AI Logic Chains, a New Breakthrough in Q&A Search
ByteDance has introduced a "Deep Thinking" reasoning mode for its AI assistant, Doubao. By visualizing the logic chain, it enhances user trust and transparency in AI. This technology, based on the Doubao 1.5 model, combined with breakthroughs in deep reasoning models, enhances the AI's intelligence and human-like qualities, suggesting broad prospects in Q&A, search, writing, and reading.
【AiBase Summary:】
🔍 The Deep Thinking mode enhances user interaction by displaying the AI's complete logic chain.
🤖 This mode is based on the Doubao 1.5 model and uses RL algorithms and engineering optimizations to enhance AI intelligence.
📈 The new feature suggests broad development prospects for AI in various fields, significantly improving user experience.
5. LTX-Video 0.9.5 Released: Commercial License Support Takes Open-Source AI Video Generation to New Heights
The release of LTX-Video 0.9.5 marks a significant advancement in open-source AI video generation technology. It not only supports commercial licenses, allowing businesses and individual developers to use the model in commercial projects, but also introduces keyframe conditional support, improving the flexibility and quality of video generation. Furthermore, the model shows significant improvements in resolution and generation speed, further meeting the needs of complex narratives.
【AiBase Summary:】
🌟 The biggest highlight is the support for commercial licenses, expanding application prospects.
🎥 The introduction of keyframe conditional support enhances video generation flexibility.
📈 Significantly improved resolution and generation speed meet the needs of complex narratives.
6. Spark-TTS Text-to-Speech System: Supports Zero-Shot Voice Cloning and Fine-Grained Control
Spark-TTS is an advanced text-to-speech system that has garnered significant attention in the AI community due to its zero-shot voice cloning and fine-grained voice control capabilities. Built on Qwen 2.5, this system simplifies the audio generation process, improves efficiency, supports multilingual generation, and is particularly suitable for audiobook production. Its BiCodec single-stream audio codec architecture ensures natural and controllable voice quality, allowing users to adjust voice characteristics as needed.
【AiBase Summary:】
🎤 Zero-shot voice cloning: Generates speaker voices without specific training data, suitable for personalized applications.
⚙️ Fine-grained voice control: Users can precisely adjust speech rate and pitch to meet different needs.
🌍 Cross-lingual generation: Supports multiple languages while maintaining high naturalness and accuracy, expanding global applicability.
Details: https://github.com/SparkAudio/Spark-TTS
7. Google Releases Preview of Whisk Animate: Transforming Images into 8-Second Animated Shorts
Google has released a preview version of Whisk Animate on its experimental AI platform, Google Labs. It allows users to leverage the advanced Veo2 model to transform static Whisk images into dynamic 8-second video clips. This new feature has quickly sparked discussions on social media, with positive user feedback demonstrating its potential in the creative industry. The launch of Whisk Animate marks a simpler and more efficient transition from static design to dynamic content, further solidifying Google's competitive advantage in generative AI.
【AiBase Summary:】
🎥 Whisk Animate uses the Veo2 model to transform static images into 8-second dynamic videos, showcasing the flexibility of animation generation.
🌟 Positive user feedback, with some early testers calling it "amazing," shows its creative potential.
🖼️ Whisk Animate provides new tools for the creative industry, simplifying the process of short video creation and advertising design.
8. Cohere Releases New Multimodal AI Model Aya Vision, Available in 32B and 8B Versions
Cohere's non-profit research lab has launched Aya Vision, a leading multimodal AI model capable of performing various language and visual tasks. Offered for free via WhatsApp, the model aims to promote technology access for global researchers. Aya Vision comes in two versions, 32B and 8B, outperforming larger competitor models. Cohere has also introduced a new benchmark evaluation tool, AyaVisionBench, to address the current evaluation crisis in the AI industry.
【AiBase Summary:】
🌟 Cohere calls the Aya Vision model industry-leading, capable of performing various language and visual tasks.
💡 Aya Vision is available in two versions, 32B and 8B, outperforming larger competitor models.
🔍 Cohere also released a new benchmark evaluation tool, AyaVisionBench, aiming to improve AI model evaluation.
Details: https://cohere.com/blog/aya-vision
9. Douyin Group Seeks AI Data Annotation Suppliers
On March 6, Douyin Group announced the recruitment of high-quality AI data annotation suppliers to meet its rapidly growing business needs. This recruitment primarily targets companies with abundant vertical resources, particularly in the medical, legal, and educational sectors. Participating companies must be independent legal entities with registered capital of no less than 1 million yuan, possess a good social credit rating, and joint applications are not accepted. This strategic move aims to enhance content quality and data service capabilities, driving industry competition and innovation.
【AiBase Summary:】
🌟 Douyin Group is recruiting AI data annotation suppliers with a minimum registered capital requirement of 1 million yuan.
📄 Applicant companies must be independent legal entities with a good social credit rating; joint applications are not accepted.
🚀 The recruitment aims to meet Douyin's rapidly growing needs in AI data annotation and to drive industry development.
10. OpenAI Launches "PhD-Level" AI Agent with a Monthly Fee of Up to $20,000
OpenAI recently announced a "PhD-level" AI agent designed to meet the high-end needs of industries such as finance, healthcare, and manufacturing. The AI agent boasts a monthly fee of up to $20,000, offering various service types with pricing based on the economic value created for clients. While the high cost has prompted some jokes, OpenAI clearly targets large enterprises, not individual users.
【AiBase Summary:】
💰 The monthly fee for this AI agent ranges from $2,000 to $20,000, with pricing based on the economic value it creates for clients.
🏢 OpenAI targets large enterprises, allowing them to pay per employee seat to lower the barrier to entry.
✈️ The AI agent aims to automate tasks with minimal human intervention, such as automatically finding flight information and completing payments.
11. Apple App Store to Introduce AI-Generated App Review Summaries, Providing Easy Access to User Feedback
Apple announced that it will introduce AI-generated app review summaries in the upcoming iOS 18.4. This feature aims to provide users with concise summaries of app reviews, helping them quickly grasp the app's highlights and key information. Summaries will be generated by a large language model and updated weekly, initially launching in the US App Store.
【AiBase Summary:】
🌟 Apple will introduce AI-generated app review summaries in iOS 18.4 to help users quickly understand app feedback.
🔄 These summaries will be updated weekly and will initially launch in the US App Store for apps with a sufficient number of English reviews.
⚠️ This feature may be susceptible to exploitation by unscrupulous businesses, affecting the authenticity and fairness of reviews.
12. IBM Launches Compact AI Model Granite 3.2, Emphasizing Efficient Inference and Practicality
IBM recently launched the Granite 3.2 large language model, focusing on providing efficient and practical AI solutions for enterprises and the open-source community. The model boasts multimodal and reasoning capabilities, enhancing flexibility and cost-effectiveness, particularly excelling in document processing and data extraction. Granite 3.2 also introduces chain-of-thought capabilities and a miniaturized secure model, Granite Guardian, ensuring high performance while reducing costs.
【AiBase Summary:】
📊 Granite 3.2 incorporates a vision-language model, enhancing document processing and data extraction capabilities.
💡 The new model features chain-of-thought capabilities, clarifying the reasoning process and enhancing reasoning ability.
🔍 The Granit Guardian security model is miniaturized by 30% without performance loss, and a verbalizable confidence risk assessment function is also introduced.
Details: https://www.ibm.com/new/announcements/ibm-granite-3-2-open-source-reasoning-and-vision
