This article gives you a quick look at this week (5.12-5.18)AI hot news
01. Major AI Events
Tsinghua University establishes the world's first AI hospital
China has established the world's first AI hospital composed of 42 AI doctors, developed by the Institute for AI Industry Research (AIR) at Tsinghua University, integrating cutting-edge technology from 21 medical departments.
In the future, it will build a closed-loop ecosystem of "AI + healthcare + education + research," promote the efficient expansion and balanced distribution of high-quality medical resources, and strive to enable more people to enjoy affordable, sustainable, high-quality medical services.
The Cyberspace Administration of China launches"Qinglang · Cracking Down on AI Technology Abuse" Phase IIcampaign
The Cyberspace Administration of China launched Phase II of the "Qinglang · Cracking Down on AI Technology Abuse" campaign.
Phase II focuses on prominent issues such as using AI technology to produce and publish rumors, false information, pornographic and vulgar content, impersonating others, and engaging in online water army activities, concentrating on cleaning up related illegal and harmful information, and handling and punishing violating accounts,MCN agenciesand website platforms.
The U.S. Department of Commerce repeals the AI Diffusion Rule signed by Biden and demands that Huawei AI chips not be used globally
The U.S. Department of Commerce announced the repeal of the AI Diffusion Rule signed by Biden, plans to introduce a more simplified rule, and adds a global ban on the use of Huawei Ascend AI chips.
The original rule's tiered system was complex and triggered opposition from tech giants. The new rule replaces tiers with bilateral agreements and strengthens supply chain controls.
This policy adjustment benefits U.S. tech companies, gives the Middle East market new opportunities for chip imports, while China continues to come under pressure.
Google releasesAlphaEvolve，can invent new algorithms, improve chip design, and solve difficult math problems
The Google DeepMind team launched a general scientific AI, AlphaEvolve, based onGeminidriven
Using two models, Gemini Flash and Gemini Pro, it continuously improves the most promising algorithms based on an evolutionary framework.
AlphaEvolve has achieved remarkable results in areas such as data center scheduling, hardware design, AI training, and solving complex mathematical problems:
- In data center scheduling, it can enable Google to recover 0.7% of computing resources;
- In hardware design, it optimized the TPU circuitry, making the chip more energy-efficient;
- AlphaEvolve improved the method of splitting matrix multiplication into subproblems, shortening Geimini's total training time by 1% and increasing its running speed by 23%;
- In tests on more than 50 open problems in mathematics, AlphaEvolve rediscovered the best solutions to date in 75% of cases. This included the "kissing number problem" that had plagued the mathematics community for 300 years, and it also overturned the 56-year-old myth of the Strassen algorithm.
From now on, AI is no longer merely solving existing problems, but can also discover new frontiers.
02. New Product Releases
Tencent Hunyuan Image 2.0 released, experience "millisecond-level" AI image generation
Tencent released the industry's first real-time image generation large model with millisecond-level response——Hunyuan Image 2.0, enabling image generation while text is being entered, with significantly improved image quality.
The world's first professional design Agent——Lovart，complete full-chain design and execution in one sentence
Lovart is the world's first professional design Agent created by LiblibAI for designers.
Lovart can think and execute design tasks like a professional designer, providing high-level design solutions.
Only one sentence is needed to complete the entire visual process from creative breakdown to professional delivery, executing hundreds of steps in a single run and producing up to 40 finished design images; images, videos, and music can be freely coordinated, integrating multiple cutting-edge AI models.
OpenAI launches Codex coding agent
OpenAI launches a cloud-based AI agentCodex, focused on automated software development, supporting parallel processing of coding, bug fixing, and other tasks.
Codex is based on the codex-1 model and is currently available as a research preview on theChatGPTplatform, with Pro, Enterprise, and Team users able to try it first.
Other Updates
QwenChatOfficially launched web development capabilities; users only need to enter a natural language instruction to generate clearly structured, aesthetically pleasing website page code, with direct preview support.
KimiReached a latest collaboration with Xiaohongshu; users can enter Kimiai conversations through the official Xiaohongshu account @Kimi智能助手 and generate notes with one click.
Kuaishou launches an AI image creation toolPoify, including features such as AI model virtual try-on, background replacement, one-click cutout, and local redrawing, empowering creativity and conversion in the e-commerce field.
OpenAI upgradesChatGPT AI chatbot，GPT-4.1as a programming expert model arrives. GPT-4.1 can precisely follow instructions and avoid lengthy output, making it more efficient, more stable in output, and better in overall performance than GPT-4o.
ManusAnnounced fully open registration, with users able to execute one task for free each day. Subsequently launched the image generation Agent feature, which can call multiple tools to work collaboratively
MiniMaxquietly launched the manus product, a general AI agent platformMiniMax Agentthat can efficiently and conveniently handle multiple tasks, such as generating audio, market research, document verification, travel planning, social media analysis, patent identification, investment strategy analysis, and creative content creation.
MiniMax releasesTTS system MiniMax Speech 02, AI voice's Her Moment, able to easily handle voices in 32 languages, different accents, and different emotions.
Qingyanlaunches an intelligent agent community, where users can directly converse with creators and post for exchanges. The app supports one-click import of WeChat files.
ChatGPTDeep Researchadds an export-to-PDF feature, which can preserve report formatting. In addition, a GitHub connector has been added, further expanding ChatGPT's application capabilities in code management and writing.
NotionLaunches AI meeting notesAI Meeting Notes, capable of real-time transcription of meeting content and automatically generating complete minutes after the meeting, including key summaries, comparison with user notes, and action items.
Tongyi Qianwenlaunches Deep Research! Based on the Qwen model, integrating multiple capabilities, it can plan complex research tasks, combine massive information from the internet, conduct multi-step search, analysis, and summarization, and generate a detailed report in just over ten minutes, with clear citation sources.
Wen XiaobaiAcademic Searchis here. The feature is powered by models such as DeepSeek R1/V3 and Qwen3, and can retrieve massive authoritative literature.
WindsurfLaunches a series of AI models designed specifically for software engineering:SWE-1, SWE-1-lite, and SWE-1-mini. It covers development stages such as code generation, testing, debugging, and information lookup, improving development efficiency and user experience.
Xindi Digitalreleases three products: Xindi Tiangong CAD 2025, Xindi Tiangong Cloud CAD 2025, and Xindi 3D Lightweight Engine 2025, deeply integrating AI technology into 3D design software.
INAIR releases a new smart product—INAIR AI Spatial Computer, planned for official release in June, equipped with a system-level AI assistant, supporting multi-window co-display and cross-device access.
03. Open Source Projects
Alibaba Open-Sources Tongyi Wanxiang Wan2.1 – VACE Model, One Model Handles Multiple Video Editing Tasks
Alibaba officially open-sources Tongyi Wanxiang "Wan2.1-VACE", supporting a full range of basic generation and editing capabilities including text-to-video, image-reference video generation, video repainting, local video editing, video background extension, and video duration extension.
Kunlun Wanwei Officially Open-Sources Matrix-Game: Building a Controllable Interactive World from Images
Kunlun Wanwei open-sourcesMatrix-Gamelarge model, the industry's first open-source 10B+ spatial intelligence large model, an interactive world foundation model specifically for game world modeling.
ByteDance Open-Sources 8B Code Model Seed-Coder, Empowering Efficient Programming
ByteDance open-sources 8B code modelSeed-Coder, including three versions: Base, Instruct, and Reasoning, respectively suitable for code completion, instruction following, and complex reasoning tasks.
Multimodal New Release, StepFun Open-Sources 3D Large Model
StepFun officially releases and open-sources 3D large model——Step1X-3D, with a total parameter count of 4.8B, composed of a 1.3B geometry module and a 3.5B texture module, capable of generating high-fidelity, controllable 3D content.
04. Investment and Financing
Meituan Solely Invests in Series A, Embodied Intelligence Company X Square Robot Has Raised Over 1 Billion
X Square Robot recently completed a Series A funding round of several hundred million yuan, led by Meituan with participation from Meituan Longzhu. In less than a year and a half since its establishment, it has completed 7 funding rounds, with cumulative financing exceeding 1 billion yuan. This funding will be used to accelerate model and robot body iteration and the implementation of intelligent solutions.
Embodied Intelligence Startup "Lingyu Intelligence" Completes Tens of Millions Seed Round Financing, Led by Inno Angel Fund
Domestic embodied intelligence company "Lingyu Intelligence" announced the completion of tens of millions of yuan in seed round financing, led by Inno Angel Fund, with participation from Shuimu Alumni Seed Fund and Vision Plus Capital. Lingyu Intelligence was founded by a team from the Department of Automation at Tsinghua University, promoting the widespread application of robots in fields such as industrial manufacturing and home services.
"OYMotion" Completes Nearly 100 Million Yuan B++ Round Financing, Second-Generation Dexterous Hand Soon to Launch
"OYMotion" recently completed nearly 100 million yuan in B++ round financing. This round was jointly invested by Infini Capital, Zhejiang Development Asset Management Co., Ltd. under Zhejiang Provincial State-owned Capital Operation Co., Ltd., and Womeida Capital, with Shengkeji Capital serving as long-term financial advisor. The funds will be used for dexterous hand technology research and development, new product launch, production capacity construction, and market expansion.
Zhongke Zidong Taichu Completes First Round of Hundreds of Millions in Financing
Zhongke Zidong Taichu (Beijing) Technology Co., Ltd. announced the completion of its first round of hundreds of millions of yuan in financing, led by CAS Star, with participation from multiple investment institutions. The funds will be used for multimodal artificial intelligence technology research and development and application, while strengthening industrial layout in the "AI+" field.Google Establishes AI Venture Capital Fund
Google established an AI venture capital fund—"AI Future Fund"—to invest in startups. Selected startups can receive investment, early access to Google's AI large models, and guidance from professional teams.
AI Artificial Intelligence ETF Receives Net Financing Buys
AI Artificial Intelligence ETF (512930.SH) received net financing buys of 7.7839 million yuan, ranking in the top tier of the entire market, with net financing buys on 3 of the last 5 trading days.
05. Insights and Perspectives
AI Reasoning Models Have the Potential for Autonomous Knowledge Generation
Jakub Pachocki, head of advanced model development at OpenAI, believes that AI reasoning models are showing the potential to autonomously generate knowledge, that AI autonomous research will make substantive progress by the end of this century, and that near-autonomous software development systems may emerge within this year.
AI investment will drive structured financing
Steven Moffat and John Greenwood of Goldman Sachs believe that as companies seek large amounts of funding for AI-related projects, this has created financing demand in the structured finance market, and more companies will benefit from this financing tool in the future.
Wikipedia unveils new AI strategy: use AI to improve volunteers' time efficiency, not replace humans
The Wikimedia Foundation has unveiled a new AI strategy, emphasizing the use of AI to assist encyclopedia editing work and improve volunteers' time efficiency, rather than replace humans. The strategy uses AI to automate tedious tasks, such as providing assistive workflow support for moderators and patrollers, improving information discoverability, automatically translating and adapting content, and scaling up onboarding training for new volunteers.Original link:AI Weekly丨AI industry developments you can't miss this week (5.12-5.18)