News

Nanoprecise recently secured $38 million in Series C funding from CIBC Innovation Banking, BMO Capital Markets, EDC, Yaletown Partners, and BDC Venture Capital.Nanoprecise is an automated predictive maintenance solutions provider, offering AI-based condition monitoring and predictive maintenance solutions that monitor machine health in a differentiated way, with meaningful insights for a variety of industries manufacturers and operations...

Recently, by the China Electric Construction Engineering Robotics Joint Innovation Center research and development of the "Zhicheng one" intelligent inspection robot made its first public appearance, the robot has been in the world's first high dam Shuangjiangkou Hydropower Station entered the testing phase, and made breakthrough progress. (Netease)

Caixun News Agency, March 30 (Xinhua), according to the release of Xiongan, on the afternoon of March 29, on the open-air soccer field outside the Xiongan Sports Center, the Yu Shu humanoid robot wearing a floral cotton dress sensitively raised its foot and kicked the soccer ball on the grass, kicking off the ball game. At the same time, wearing a set of lion dance costume Yu Shu robot dog running around outside the field, stopping from time to time to the field of small players waved, than a heart, for the small players to cheer. These two robots are Yu Shu robot debut in Xiong'an. (Sohu)

According to The Verge, Microsoft CEO Satya Nadella recently said in an internal all-hands meeting that DeepSeek's R1 model has become a new benchmark for Microsoft's AI development. In particular, Nadella emphasized the efficient operation of the DeepSeek team: "What's most impressive about DeepSeek is that it shows what a team of 200 people working together can accomplish. What's more, they didn't just stop at a research project or an open source project, they built it into the #1 product in the app store. This is what I have in mind...

On March 27, Artificial Analysis, a world-renowned AI benchmarking organization, released the latest global video generation big model list, fast hand can spirit 1.6pro (high-quality mode) with 1,000 points of Arena ELO benchmarking scores landed on the top of the Image to Video (Image to Video) track, Google Veo 2, Google Veo 2 and Pika Art were second and third on the list. In December of last year, Kerin launched the 1.6 model, which was recognized for its video-generated text responsiveness, dynamic...

On March 30th, 2025 China Science Fiction Convention was recently held in Beijing, with the theme of "Scientific Dreams Creating the Future". In the forum of the conference, science fiction writer Liu Cixin talked about the drive and development of DeepSeek on science fiction literature. According to CCTV Finance, on the question of whether DeepSeek is likely to replace science fiction writers in the future, Liu Cixin said that it would be unlikely for the time being, but after 10 or 20 years, it is theoretically possible to replace science fiction writers. According to Liu Cixin, from a scientific point of view, there is nothing in all human writers that cannot be replaced by AI...

OpenAI upgraded ChatGPT's text-to-picture function to the GPT-4o model on March 25, claiming to be able to more accurately generate pictures based on users' text commands, and to quickly convert user-uploaded images into manga, anime, and other styles. The corresponding function triggered the trend of "AI generating Studio Ghibli-style pictures" on overseas social media, with many people trying to convert their selfies into the corresponding style, and even OpenAI CEO Sam Altman changing his X platform avatar to the "Ghibli version" at one point. However, pressure on the feature...

On March 30, Gurman broke the news in the Power On newsletter that Apple plans to expand its Health app into an intelligent health assistant that will continue to collect data provided by users' devices and make personalized health improvement recommendations based on that data through an AI coach. To that end, Apple will build a new facility near Oakland, California for doctors to shoot video content. The company is also looking for a well-known doctor to serve as a host for the new service, which is currently being tentatively called 'Health+' within Apple.

On March 30, Google announced that the latest Gemini AI flagship model, Gemini 2.5 Pro, will be free to all Gemini app users. Previously, the experimental model was limited to Gemini Advanced subscribers. Google released Gemini 2.5 Pro earlier this week, calling it "the smartest AI model" to date, with further enhancements in reasoning capabilities over its predecessor. The new version of the model supports a number of features, including app and browser extensions, file uploads, and integration with Google ...

On March 29, at the 2025 Zhongguancun Forum Annual Meeting of Artificial Intelligence Leading Talent Development Forum, Haidian District launched the "Zhongguancun AI Beiwei Community", in which the enterprises can enjoy rent reduction, industrial policies, talent support and enterprise service empowerment, and all-around support for the development of AI startups. For qualified AI startups, a three-year lease contract can be signed to provide three types of support: high-level universities, research institutes and new research and development institutions and other scientific and technological achievements transformation projects for a maximum of three years rent-free; the first two years of the newly-added enterprises rent-free, and the third year is reduced by half; the first year of the enterprises in the region rent-free, and the second year is reduced by half. In addition, the neighboring ...

In response to the online discussion about the beginning and end of the cooperation between GSR Ventures and StarHub, the embodied intelligent robotics company StarHub released a statement saying that GSR Ventures participated in the company's angel round of financing in November 2023, and put forward a request to the company to withdraw from the equity investment in May 2024, and the company cooperated with the company positively, and assisted in the successful completion of the withdrawal of the equity investment in September 2024, the company was informed. The Company has never had any direct communication with Mr. Zhu Xiaohu, nor does it have his contact information, and the relevant rumors are untrue, "We deeply respect Mr. Zhu's investment achievements and industry contributions."

Recently, lottery players buy lottery prediction software received 100 express cheated triggered concern, the reporter interviewed the staff of the lottery, the staff, the lottery each draw is an independent random event, not subject to the influence of the previous results, and there is no law, so the AI can not predict the results of the lottery winners. In addition, the China Fortune Lottery Center has never carried out Internet sales of lottery activities through any platforms, self media, web pages, etc., nor has it carried out Internet sales of lottery activities through cooperation with any units and individuals; please be vigilant against individuals, organizations, and online platform accounts claiming to be able to predict the lottery winning results, to prevent being fooled...

The OWL team used the CAMEL-AI framework to replicate Manus in one day, and the measured performance reached the open source community's GAIA performance of 58.18%, surpassing OpenDeepResearch's 55.15%; there are technical differences between Manus and OWL, but both use multi-agent systems to accomplish tasks, and OWL believes that MCP is a future trend that will enable frameworks to access a unified standard tool; Generalized Agent and Vertical Domain Agent each have their own advantages, and for the vertical domain, the key is to solve the core pain points of a specific domain and provide professional UI/...

Anthropic develops an AI "microscope" method, which reveals for the first time the thinking process and information flow paths of the large model Claude by tracking the internal activity patterns of the neural network; the study finds that Claude plans its output in advance, and has the ability to share concepts in multiple languages, parallel computational paths, and multistep inference, rather than simply generating words word by word; the team reveals the internal mechanism of Claude when dealing with "hallucinations" and refusing to answer questions and encountering jailbreak attacks by intervening in experiments, providing a new method to improve the reliability of AI. Through intervention experiments, the team reveals Claude's internal mechanisms for dealing with "hallucinations", refusing to answer, and experiencing jailbreak attacks, providing a new way to improve AI reliability.

The ECgMPL model can accurately identify cancer from cellular and tissue microscopic images, and the diagnostic accuracy of endometrial cancer reaches 99.26%, which is much higher than the average of 78-81% for doctors; the model is adjusted to be versatile and can diagnose a wide range of cancers, and the accuracy of colorectal cancer is up to 98.57%, breast cancer 98.20%, oral cancer 97.34%; the research team used multi-step image processing technology and innovative ECgMLP architecture to position the model as a diagnostic aid for doctors, with the advantages of speed and low cost.

Musk announced the sale of its social platform X to its own AI company xAI in an all-stock transaction, after the merger of the new company valuation of 113 billion U.S. dollars; the merger reflects Musk's "All in AI" strategy, X valuation of 33 billion U.S. dollars (45 billion minus 12 billion debt), xAI valuation of 80 billion U.S. dollars; the acquisition of the integration of the data, modeling and computational resources, X platform of user data will provide xAI with a rich training corpus, the new company valuation exceeded Musk's previous acquisition of OpenAI offer. User data will provide a rich training corpus for xAI, and the new company's valuation exceeds Musk's previous acquisition offer for OpenAI.

VAST will open source 8 major 3D generation projects in one breath, of which TripoSG and TripoSF have been the first to be released; TripoSG as the basic 3D generation model, using the corrected flow Transformer architecture, the number of parameters up to 4 billion, able to generate high-detailed 3D mesh models from a single image; TripoSF introduced the SparseFlex representation method TripoSF introduces the SparseFlex representation, which supports high-resolution (up to 1024³) mesh reconstruction with arbitrary topology, low memory footprint and end-to-end training through rendering loss.

Tencent Yuanbao is upgraded to recognize and process 36 file formats, including Word, PDF, Excel and various types of development files (.py, .java, .json, etc.); in development scenarios, Yuanbao is able to complete tasks such as code interpretation, review, bug location, modification suggestions, syntax optimization, and language conversion; this upgrade is the third feature iteration for development scenarios in a week following the access to DeepSeek V3-0324 This upgrade is the third feature iteration of Yuanbao for development scenarios in a week after accessing the latest version of DeepSeek V3-0324 and supporting real-time preview of HTML code.

Facade Intelligence launched cpmGO, the first pure end-side intelligent assistant, emphasizing that the end-side model is different from the end-side AI deployed in the cloud, and has the advantages of privacy, security and offline operation; the super assistant of the small steel cannon has landed in the field of automobile intelligent cockpit, realizing the ability of full-chain perception and decision-making execution, and supporting 9 categories and 26 applications, and its partners have occupied the three seats of the intelligent automobile sales list; Facade puts forward the concept of "model is Agent" and the "law of density", believing that the development of large models will be the trend in the future. The concept of "Model is Agent" and "Law of Density", that large models will become smaller and smaller but stronger and stronger, and the future development of end-side models is the general trend.

GPT-4o found a new function "Brush Edit", which supports one-click keying, changing clothes and background, and users can precisely modify the image by smearing the area and inputting commands; multiple user tests show that the function has strong ability to retain details, and it can complete the operations of changing clothes, adding accessories, and replacing the background, etc., but there is room for improvement in the effect of complex transformations; GPT-4o shows signs of inference ability, and users found that the model can display the inference time and thought chain process, indicating that OpenAI is merging inference and non-inference models or preparing for GPT-5. GPT-4o shows signs of reasoning ability, and users found that the model can display the reasoning time and the chain of thought process, which indicates that OpenAI is merging reasoning and non-reasoning models, or preparing for GPT-5.