Recent developments in AI span a wide range of applications and geopolitical implications. Huawei's CloudMatrix 384, powered by Ascend chips, has shown competitive performance against Nvidia's H800 in certain AI tasks, specifically running DeepSeek's reasoning model, though it consumes significantly more power. Despite US sanctions, Huawei plans to ship 700,000 Ascend AI processors by 2025, signaling China's push for AI independence. The White House has warned that China is closing the AI gap with the U.S., driven in part by trade regulations. In the realm of AI safety and governance, China has cracked down on over 3,500 AI products for breaking regulations, while experts emphasize the need for robust AI governance frameworks to manage risks. On the consumer front, Applebee's and IHOP are integrating AI for customer service, and AI avatars are generating millions in sales in China, with avatars of Luo Yonghao and Xiao Mu generating $7.65 million in six hours on Baidu's Youxuan. Meanwhile, companies like Veesion are using AI to prevent theft in retail stores by analyzing body movements. However, challenges remain, as Tesla's former AI head, Andrej Karpathy, points out that self-driving technology is not yet fully solved, and Google's new AI conversation feature has usability and accuracy issues. OhChat is creating AI versions of celebrities using Meta's language model, and Pope Leo XIV has called for an ethical AI framework.
Key Takeaways
- Huawei's CloudMatrix 384 outperformed Nvidia's H800 in a DeepSeek R1 LLM test but uses four times as much power.
- Huawei plans to ship 700,000 Ascend AI processors in 2025 despite US sanctions.
- The White House warns that China is closing the AI gap with the U.S.
- Veesion's AI is used in 5,000 stores to detect shoplifting by analyzing body movements.
- Applebee's and IHOP are using AI for customer service and personalization.
- OhChat creates AI versions of celebrities using Meta's language model, sharing 80% of revenue with creators.
- AI avatars of Luo Yonghao and Xiao Mu generated $7.65 million in six hours on Baidu's Youxuan.
- Tesla's former AI head says self-driving technology is not yet fully solved.
- China has cracked down on over 3,500 AI products for breaking rules.
- Pope Leo XIV calls for an ethical AI framework.
Huawei's AI beats Nvidia in one test but uses more power
Huawei's CloudMatrix 384 AI cluster, using Ascend 910C chips, outperformed Nvidia's H800 chip in a DeepSeek R1 LLM test. The CloudMatrix combines 384 NPUs and 192 CPUs for fast communication. While it delivers more raw power than Nvidia's GB200 NVL72, it consumes four times as much power. This makes it a compelling option for Chinese customers needing peak LLM performance, especially with limited access to Nvidia products.
Huawei's AI chip strategy overcomes US sanctions, boosts China's tech goals
Despite US sanctions, Huawei is set to ship 700,000 Ascend AI processors in 2025. Huawei's CloudMatrix 384, powered by Ascend chips, outperformed Nvidia's GPUs in running DeepSeek's reasoning model. The system uses 384 Ascend 910C NPUs and 192 Kunpeng CPUs. Huawei's CEO says that using methods like stacking and clustering resulted in computing performance comparable to the most advanced systems in the world. US export controls have pushed China to develop its own AI chipmaking industry.
White House warns China closing AI gap with the U.S.
The White House warns that Chinese tech firms like Huawei are catching up to U.S. AI capabilities faster than expected. David Sacks says strict U.S. trade regulations are pushing countries to use Chinese products. He estimates China is now only about two years behind the U.S. in chip design. Sacks also mentioned the end of a Biden-era policy that would have made it harder to export high-end AI chips.
AI security watches store gestures to help prevent theft
Veesion, a Paris-based startup, uses AI to help stores prevent theft. Their system analyzes body movements from security cameras to detect suspicious gestures. The AI flags actions like hiding items, without tracking faces or identities. This technology is used in 5,000 stores across the U.S., Canada, and Europe to reduce losses.
AI security spots suspicious behavior to stop shoplifting
Veesion, an AI startup, developed an algorithm that recognizes gestures to predict shoplifting. The tech flags suspicious movements to store security, without focusing on appearances. Veesion's system is used in 5,000 stores across Europe, Canada, and the US. One store cut missing inventory in half using Veesion's technology. The AI analyzes camera footage to detect movements and objects, alerting staff to potential theft.
Applebee's and IHOP add AI for support and better customer service
Applebee's and IHOP will use AI tools for customer service and behind-the-scenes tasks. Dine Brands, the parent company, is rolling out AI to 300 franchisees and 3,500 stores. One tool is AI-powered tech support that uses natural language to answer questions. They are also adding a 'personalization engine' that uses past purchases to suggest items and promotions. IHOP partnered with Google Cloud in 2023 to use AI for online ordering.
OhChat creates AI versions of celebrities like OnlyFans
OhChat is a new platform that uses AI to create digital twins of public figures. Users can pay to interact with these AI avatars, like Katie Price's digital twin, Jordan. The platform has 200,000 users and offers subscriptions for texts, voice notes, and images. Creators receive 80% of the revenue their AI avatar generates. OhChat uses Meta's language model to generate the digital replicas.
AI avatars drive millions in sales on China's YouTube
AI avatars are boosting livestream shopping sales in China. These AI hosts mimic real people and stream 24/7 on platforms like Douyin and Baidu. On Baidu's Youxuan, AI avatars of Luo Yonghao and Xiao Mu made $7.65 million in six hours. These avatars are trained on video footage to copy voices and styles. AI hosts don't need breaks and cost less, making them popular for online sales.
Tesla's former AI head says self-driving cars not yet solved
Andrej Karpathy, Tesla's former head of AI, warns that self-driving technology is not fully solved. He notes that even Waymo, which operates autonomous rides, still uses teleoperation. Karpathy believes AI agents, like those driving vehicles, will take time to develop. He says that while progress is being made, true level 4 autonomous systems are still years away.
China cracks down on 3,500 AI products for breaking rules
China has taken action against over 3,500 AI products since April for not following rules. The products include mini-programs, web apps, and AI agents. The crackdown targets misuse of AI, like deepfakes, and failure to label AI-generated content. China removed over 960,000 items with illegal content and shut down 3,700 accounts. The Cyberspace Administration of China will focus on AI-generated rumors and online content in the future.
AI Governance framework helps security leaders manage AI risks
AI is being used in many business areas, but most companies lack a formal governance structure. This can lead to problems like flawed outputs and security issues. An AI governance framework helps make AI safer and more transparent. Key principles include accountability, transparency, fairness, resilience, and privacy. Classifying AI risk levels and implementing security controls are also important for a robust framework.
Google's new AI conversation feature is not convenient
Google is testing a new conversation feature for AI Mode in its search app. This feature allows users to ask questions out loud and receive spoken answers from AI. However, the feature lacks a pause button and doesn't easily allow for repetition of information. While the live sources feed is useful, the AI can cite unreliable sources like random Facebook posts. The accuracy of the AI is also a concern, as Google admits it can make mistakes.
Pope Leo XIV calls for ethical AI framework
Pope Leo XIV urged AI companies to create an ethical framework. He said AI needs to consider the well-being of humans in material, intellectual, and spiritual ways.
Sources
- Huawei's brute force AI tactic seems to be working — CloudMatrix 384 claimed to outperform Nvidia processors running DeepSeek R1
- How Huawei’s silicon strategy defies US sanctions, advances China’s AI ambitions
- White House crypto and AI adviser warns China catching up to the U.S. — 'Before DeepSeek, people thought that Chinese AI models were years behind and we realized that they are only months behind'
- New AI Security Tech Flags Suspicious Gestures to Help Stores Prevent Theft
- This AI security tech alerts store staff if it thinks you're trying to steal something
- Applebee’s and IHOP to Deploy AI-Powered Tech Support and Personalization
- ‘She never sleeps’: This platform wants to be OnlyFans for the AI era
- How AI avatars are driving million-dollar sales on China’s version of YouTube?
- Tesla's former head of AI warns against believing that self-driving is solved
- China tackles over 3,500 non-compliant AI products since April
- AI Governance Framework: For Security Leaders
- I Tried Google's Conversation Feature for AI Mode, and I'm Not Sure Who It's For
- Pope Leo XIV urge A.I. companies to create ethical framework