24-hour AI news: Tongyi Qianwen open-sources new multimodal technology, and Sora introduces copyright protections.

In the past 24 hours, domestically, there has been a simultaneous development of model open source and industrial implementation, as well as AI governance and rule of law issues; overseas, OpenAI Sora has triggered a dual focus on copyright and content security, while NVIDIA and Fujitsu have expanded their strategic cooperation.

1. Alibaba Cloud Tongyi Qianwen Open Source Qwen3-VL-30B-A3B Multimodal Model

1. On October 4, it was announced that Instruct and Thinking would be open source, with FP8 versions available. It emphasized that "only about 3 billion activation parameters" can cover VQA, OCR, video understanding, and intelligent agent tasks.
2. Simultaneously launch the FP8 version of Qwen3-VL-235B-A22B, targeting high-end visual language scenarios.
3. It was disclosed that some evaluations were benchmarked against GPT-5-Mini and Claude 4-Sonnet, with the intention of expanding industry implementation through an open source ecosystem.

2. The latest progress of Beijing Foreign Studies University's "Digital Virtual Teacher" program

1. The laboratory released a "digital virtual teacher" system that supports multilingual and cross-cultural responses, emphasizing more rigorous reasoning and localized pragmatics.
2. Focus on language intelligence and educational scenarios, taking into account multilingual teaching and cultural understanding.
3. As a sample of AI applications in universities, expand the implementation path of "AI+Education".

3. Huashengchang released four "DeepSense large models + AI medical devices" at the Guangzhou Medical Equipment Exhibition

1. Launched new products including ECG blood pressure monitor, home portable six-lead ECG, bi-level ventilator, smart blood pressure monitor, etc., all equipped with self-developed DeepSense model.
2. DeepSense has completed the registration of its generative AI service and is in the preparation stage for open internal testing.
3. Focus on the "AI+Medical" vertical large-scale model route and explore the commercialization of equipment-model integration.

4. Beijing Hosts the "Global Rule of Law Conference in the Era of Artificial Intelligence"

1. The conference was held at Renmin University of China from October 3rd to 4th, with approximately 400 guests attending from more than 30 countries and regions.
2. Set up multiple sub-forums around topics such as AI supervision, platform governance, and legal education.
3. Strengthen collaborative research between China’s solutions and the legal community in the global AI governance process.

5. Policy and Industry Observation: “AI+” Action Continues to Heat Up

1. Mainstream media focused on the practical application of "artificial intelligence +" and showcased new scenarios such as exoskeleton robots.
2. Previously, relevant departments proposed to study and formulate an implementation plan for the "Artificial Intelligence + Manufacturing" special action.

6. OpenAI announces new measures regarding Sora copyright and commercialization

1. Provide a “shieldable” and authorized channel for the rights holder role and IP, and explore the profit-sharing mechanism for the permitting party.
2. Sora has been launched in the United States and Canada as an independent app, focusing on 10-second AI video generation and "guest appearance" functions.
3. As copyright and profit-sharing plans are still in the pilot stage, industry competition and standardization paths have become the focus.

7. Sora topped the App Store's US free chart, sparking controversy over content safety.

1. It topped the US free list within a few days of going online, and invitation codes were resold in the secondary market.
2. The media pointed out that examples of violence, hatred and other problems appeared on the platform, questioning the "effectiveness of the guardrails."
3. Discussions surrounding content review, protection of minors and the boundaries of copyright exemptions are heating up.

8. NVIDIA and Fujitsu expand their collaboration to build a "full-stack AI infrastructure" and industry AI agent platform

1. An integrated computing power base will be built using FUJITSU-MONAKA series CPUs and NVIDIA GPUs through NVLink Fusion.
2. Focus on implementing AI agent platforms in manufacturing, medical care, robotics and other fields, positioning enterprise-level transformation.
3. Both parties envision establishing large-scale AI infrastructure and application capabilities in Japan by around 2030.

9. Bezos at Italian Tech Week: AI is in an "industry bubble," but its long-term value is certain

1. The "industrial bubble" theory was put forward: capital overheating and good and bad ideas coexist, but the technological dividend is real.
2. In response to the market’s divergent views on AI valuations, it reinforces the judgment of “short-term fluctuations and long-term positive sum”.
3. The linkage between the AI cycle and data centers, hardware, and application layers remains optimistic.

10. Google Photos tests using AI to quickly generate videos from photos

1. Foreign media reported that Google will enhance the AI generation function of Photos and support automatic generation of videos from albums.
2. The line of lightweight generative tools for mass creation continues to expand.
3. The competition with short videos and AI video applications has further intensified.

Frequently Asked Questions (Q&A)

Q: Why did Sora quickly spark controversy after its launch?

A: On the one hand, its app reached the top of the US app rankings, experiencing a surge in traffic. On the other hand, media outlets have highlighted instances of inappropriate content, including violence and hate speech. OpenAI has also proposed offering rights holders IP blocking and a pilot revenue-sharing mechanism to alleviate copyright and censorship pressures.

Q: What are the key indicators of Alibaba Qwen3-VL-30B-A3B?

A: Official information shows that two versions, Instruct and Thinking, and an FP8 inference version are available. It is claimed that it achieves cost-effective performance with "approximately 3 billion activation parameters" in tasks such as STEM, VQA, OCR, video understanding, and intelligent agents. A larger model FP8 version will be released simultaneously to cover heavy-load scenarios.

Q: What are the technical highlights and timeline of the NVIDIA-Fujitsu collaboration?

A: We will use NVLink Fusion interconnect to integrate GPUs and FUJITSU-MONAKA CPUs, and provide a supporting industry AI agent platform, with priority given to manufacturing/medical care/robotics. Our goal is to establish a full-stack AI infrastructure and application ecosystem in Japan for 2030.

Q: What signals does the recent progress of “artificial intelligence+” in China send?

A: Mainstream media have continuously focused on the scenario-based implementation of AI in consumption, people's livelihood and industry; the policy side has previously proposed the "artificial intelligence + manufacturing" special action direction, releasing a dual-wheel drive signal from policy to application.