Search AI Products and News

Google's NotebookLM introduces video summaries, converting complex info into narrated slides. It auto-generates videos from user-uploaded images/charts. Demoed at I/O 2025, the feature will roll out gradually. Currently English-only, multilingual support planned.....

OpenAI launches ChatGPT Study with interactive prompts and scaffolding responses for systematic learning across subjects. Available to all users, with an Edu version coming soon. Educators see potential to transform teaching methods.....

Doubao App has made a major upgrade in the field of visual reasoning. Its image analysis feature now supports a deep thinking mode, offering users an unprecedented intelligent experience. Users just need to take or upload a picture while in deep thinking mode, and Doubao can quickly perform fine adjustments such as zooming and cropping, and supports image search functionality, enabling users to think and search simultaneously, thereby further improving the accuracy of search results.

Step Zenith has officially launched its new research assistant, 'Step Deep Research,' which is now available for experience on the Step Zenith App and stepfun.com platform, while also opening internal invitation testing. This intelligent tool is designed for fields such as finance, consulting, healthcare, law, policy, and academic research. It can independently complete complex research tasks in about ten minutes and generate reports that are both in-depth and professional, providing efficient support to researchers. The core advantage of 'Step Deep Research' lies in its powerful information processing and integration capabilities. In the context of Sequoia China

Perplexity company launches an AI browser called Comet, which transforms YouTube videos into interactive learning tools. Users can pause videos at any time to ask questions, and the AI assistant provides instant answers and supplements related knowledge, enabling a shift from passive watching to active learning. This tool is especially suitable for complex subjects, as it optimizes answers based on user habits and offers personalized learning paths. Its deep integration with the browser environment makes responses more contextually relevant. Although still in the testing phase, it has already shown the potential to revolutionize the way education is delivered.

VLM2Vec-V2, a multimodal embedding framework developed by Salesforce Research and other institutions, breaks through the limitations of traditional models. Based on the Qwen2-VL architecture, this framework innovatively unifies image, video, and visual document retrieval tasks. It adds five new evaluation tasks to expand the MMEB dataset. With key technologies such as dynamic resolution and M-RoPE, it achieved an average score of 58.0 across 78 datasets, leading in performance, especially in video tasks. Although its document retrieval is slightly behind ColPali, it

iFlytek launches Spark X1 upgrade on July 25, a Chinese AI model excelling in math, translation, and logic with fewer parameters but comparable to OpenAI/DeepSeek. Features improved accuracy, multilingual support (130+ languages), and precise responses via 'slow thinking' mode, marking a breakthrough in domestic AI tech.....

WordPecker is an open-source AI language learning tool featuring personalized vocab extraction, 100+ languages, voice chat with OpenAI Agent, and interactive exercises. It represents the future of smart language learning.....

Mistral AI's chatbot Le Chat has received a major update with five core features now available: 1) Deep Research mode that intelligently breaks down complex questions and generates structured reports; 2) Voice input feature based on the Voxtral model for natural conversation; 3) Thinking mode using the Magistral model to handle complex reasoning; 4) A text-to-image modification feature developed in collaboration with the Black Forest Lab; 5) New project management tools to organize conversations and files. These features are now available on both web and mobile platforms, significantly enhancing AI capabilities.

Recently, the ByteDance Seed team collaborated with the University of Hong Kong and Fudan University to introduce an innovative reinforcement learning training method called POLARIS. This method successfully enhances the mathematical reasoning capabilities of small models to levels comparable to those of large models through a carefully designed Scaling RL strategy, offering a new approach for optimizing small models in the field of artificial intelligence. Experimental results show that the 4 billion parameter open-source model Qwen3-4B trained using POLARIS achieved remarkable performance on AIME25 and AIME24 mathematical tests.

On July 15, Mita AI Search announced the completion of a new iteration of its 'Deep Research' module and its official public beta launch, becoming the first search service in China that is freely open to the public and features multi-turn reasoning chain visualization. The upgraded system adopts a segmented reinforcement learning strategy, breaking down the originally computationally intensive 'Deep Research' tasks into multiple sub-tasks, maintaining result accuracy while reducing operational costs to a level that allows for public free access. It performs particularly well in retrieving and reasoning with Chinese corpus.

Microsoft has released the public preview of Deep Research, a new service called Azure AI Foundry. This service functions as a research assistant similar to OpenAI agents. It can automatically break down complex tasks, perform multi-turn information retrieval and verification using Bing search and GPT models, and generate audit-able research reports. The service is applicable to fields such as academia, finance, and healthcare, and supports API integration, significantly improving research efficiency. Applications are now open, and developers can integrate its automation capabilities into their own applications.

Microsoft launches the Deep Research research tool, which integrates Bing search and OpenAI technology to automate research. The tool uses the core technology o3-deep-research, with a workflow that includes four key steps: first, interacting with GPT-4o/4.1 to clarify user requirements; second, calling Bing to retrieve the latest data; third, performing intelligent analysis and reasoning; finally, generating a structured report containing answers, reasoning process, cited sources, and clarification records. The tool supports integration with Azure AI

Google recently introduced a new strategy that allows the Gemini AI assistant to access third-party applications such as WhatsApp, raising privacy concerns. Even if users disable interaction permissions, Gemini can still access some application data, which will be reviewed by humans and retained for 72 hours. Although an option to disable it is provided, the instructions are unclear, making it difficult for users to confirm whether the permission has been successfully disabled. Experts have criticized this strategy as similar to Microsoft's forced bundling of IE, and suggest uninstalling Gemini via the command line, although the process is complex. This incident highlights the challenges faced by tech companies.

Large Language Models (LLMs) have achieved significant progress in complex reasoning tasks by combining task prompts with large-scale reinforcement learning (RL), as demonstrated by models like Deepseek-R1-Zero, which directly apply reinforcement learning to base models, showcasing strong reasoning capabilities. However, this success is difficult to replicate across different base model families, especially within the Llama series. This raises a core question: what factors lead to inconsistent performance of different base models during reinforcement learning? How does reinforcement learning perform in

In a recent major announcement, the global e-commerce and cloud computing giant Amazon revealed two important milestones in its robotics and artificial intelligence (AI) fields: the launch of a new AI foundation large model called Deep Fleet, and the successful deployment of more than one million robots. The launch of the Deep Fleet model aims to enhance the intelligence and efficiency of Amazon's largest industrial mobile robot fleet. The application of this model is expected to improve the operational efficiency of the robot fleet by 10%, thereby accelerating...

OpenAI announced the official opening of API access to its deep research models, providing developers with a powerful set of tools including automatic web search, data analysis, MCP (Model Communication Protocol), and code execution. The models opened include the deep research versions of o3 and o4-mini, which have already been used in ChatGPT. Now, developers can directly call these models via API. These models are particularly suitable for complex tasks that require obtaining the latest information and performing advanced reasoning. In terms of function expansion, o3,

New Oriental officially launched its first consumer-facing original AI education product - New Oriental AI 1-on-1 today. This is not only a major breakthrough in teaching methods, but also marks a critical step in New Oriental's strategic layout of "Education + AI." The core competitiveness of New Oriental AI 1-on-1 lies in providing learners with a high-frequency interactive 1-on-1 learning experience. The AI teacher can realistically reproduce the learning environment, achieving real interaction and real Q&A. At the same time, the AI teacher is patient, responsible, proficient in teaching, and can provide timely feedback, as well as praise and encourage students.