Welcome to the [AI Daily] column! This is your daily guide to exploring the world of artificial intelligence. Every day, we bring you hot topics in the AI field, focusing on developers and helping you understand technology trends and innovative AI product applications.
Fresh AI products, click to learn more: https://app.aibase.com/zh
1. Baidu releases its first physical AI world model Apollo World; nearly 100 Apollo Go robotaxis operate with the new architecture
Baidu Apollo World models traffic participants as autonomous agents and can conduct about 100,000 kilometers of virtual road testing per day to verify the safety of end-to-end architectures, reflecting L4's evolution from real-vehicle testing to a simulation closed loop. However, world models still struggle to fully simulate real environments and need to be combined with real-vehicle data.
[AiBase Highlights:]
🌍 Baidu launches its first physical AI world model Apollo World, strengthening end-to-end safety verification
🚗 Nearly 100 Apollo Go robotaxis are equipped with the new end-to-end architecture, with cumulative safe mileage approaching one million
⚠️ World models still cannot fully simulate real environments and need to be combined with actual road data
2. Tencent Hy4 official version debuts: entering the strongest open-source camp, focused on real work
The official version of Tencent Hy4 achieves breakthroughs in vision and computer operation, remaining open source while keeping its price unchanged, but it admits there is still a gap with top closed-source models, showing a pragmatic iterative posture.
[AiBase Highlights:]
👁️🗨️ Introduces vision and computer operation for the first time, supporting command line and graphical interfaces
💰 Performance exceeds preview but API price remains unchanged, now available on platforms such as Tencent Cloud
⚡ Admits a gap with top closed-source models; a cost-effective flash version will be launched soon
3. Alibaba Cloud releases AI investment research workbench Qovest: open personal invitation testing at 99 yuan/month
This product packages financial data, agent frameworks, and investment research workflows into a standardized SaaS, targeting professional investors' rigid demand for accurate data and traceable conclusions; its self-developed Harness and leading evaluation are highlights, and the personal version is priced relatively low. Whether it can break through on the institutional side depends on the depth of the enterprise version's workflows.
[AiBase Highlights:]
🤖 A one-sentence request can automatically retrieve data, conduct in-depth analysis, and generate reports
🛠️ The self-developed Harness framework connects the entire process from data retrieval to analysis output
📈 The personal version is open from 99 yuan/month, and an enterprise version will be launched, supporting dynamic tracking dashboards
4. Reports say ByteDance broadly restricts employees from using third-party AI tools at their own expense
ByteDance is tightening control over external AI tools, with the core aim of preventing leaks of L4-level code and document data and protecting core assets. This move reflects big tech's strong control over self-developed ecosystems and data compliance, but it may also affect R&D efficiency and freedom of tool choice.
[AiBase Highlights:]
🔒 Fully bans external AI access to L4-level code repositories and Feishu documents
⚠️ The core motivation is to prevent data leaks and asset security risks
🧠 Reflects big tech accelerating self-developed ecosystems and strengthening data compliance controls
5. European Court adviser warns: Gemini may be infringing if it bypasses paywalls or reuses content
The EU Advocate General's opinion clarifies the boundaries of AI copyright: facts are not protected, but bypassing paywalls or identifiably reusing news expression may constitute infringement. If the court adopts this, it may push platforms and publishers to establish a fairer mechanism, benefiting traditional media.
[AiBase Highlights:]
⚖️ Pure facts are not protected by copyright, but forms of expression are protected
🤖 Training models is a lawful exception, but bypassing paywalls may constitute infringement
📰 If Google loses, traditional media may gain a turning point for rights protection and revenue sharing
6. One person can get it done in a day: the AI copilot model behind Windows 11's native speed test tool
Microsoft demonstrates an efficient human-machine collaborative development path: one person completed a native speed test tool prototype in spare time with the help of an AI agent and integrated WinUI 3. The core is quickly turning concepts into usable value while emphasizing quality bottom lines, rejecting AI junk, and providing a reusable paradigm for internal AI-assisted development.
[AiBase Highlights:]
🛠️ An employee used an AI agent to build a native speed test prototype in one day in spare time
🪟 Upgraded from a Bing web wrapper to a WinUI 3 native tool
🎯 Humans set direction, AI executes, emphasizing delivery of real value rather than AI junk
7. Apple AI accelerates entry into the Chinese market, with Siri and Visual Intelligence integrating Qwen
Apple has opened the first batch of Apple Intelligence features to mainland China users through the iOS 27.3 beta, integrating Alibaba Qwen and Baidu Search, showing accelerated localization compliance and ecosystem cooperation. The first batch of features is relatively basic; Siri calls Qwen, but deep personalization requires redirection. The new Siri has not yet arrived, and the experience remains to be verified.
[AiBase Highlights:]
📱 The mainland China iOS 27.3 beta opens the first batch of Apple Intelligence features
🤖 Siri and Visual Intelligence integrate Alibaba Qwen and Baidu Search
📞 Real-time translation supports Messages, Phone, and FaceTime, as well as specific models
8. China's first Sanjin culture large model "Jinwen AI" officially launches, with digital human "Jindou" unveiled at the same time
This model uses "LLM+RAG+Agent" to integrate publication-grade authoritative corpus, engineering constraints on hallucination and latency, and has relatively high knowledge credibility; seven major agents and AR-guided tours form a clear implementation path. The key going forward is continuous content updates and large-scale replication across scenarios.
[AiBase Highlights:]
🧠 Adopts an LLM+RAG+Agent architecture, with a hallucination rate below 1% in the authoritative knowledge base
📚 Seven major Shanxi culture agents cover scenarios such as ancient architecture, Shanxi merchants, and cultural creativity
🥽 AI+AR intelligent guided tours have been implemented at the Shanxi Merchants Museum and received multiple industry recognitions
