• AI KATANA
  • Posts
  • Tencent testing DeepSeek to add China’s hottest AI technology to its Weixin super app

Tencent testing DeepSeek to add China’s hottest AI technology to its Weixin super app

Also: Meet Tanka, the AI Messenger with Memory for Teams

Author

AI KATANA
February 17, 2025

Hello!

AI adoption continues to accelerate across industries, with major developments in China, South Korea, and beyond. Tencent is testing DeepSeek AI in its Weixin super app, following a broader trend of Chinese tech giants integrating domestic AI models. Meanwhile, a Microsoft and Carnegie Mellon study highlights concerns over AI’s impact on critical thinking, urging better verification mechanisms. In healthcare, the University of Hong Kong has introduced CytoMAD, an AI-powered tool for faster and more precise cancer detection. On the global AI infrastructure front, South Korea is securing 10,000 GPUs to strengthen its national AI computing center while also banning China’s DeepSeek chatbot over data security concerns. The New York Times is deepening its investment in AI-driven newsroom tools, reflecting the media industry’s cautious yet growing embrace of automation. In venture capital, Ilya Sutskever’s Safe Superintelligence is reportedly raising over $1 billion at a $30 billion valuation, emphasizing AI safety, while UK-based Ori Industries secured $140 million to expand access to Nvidia superchips. These developments showcase the ongoing global race in AI infrastructure, governance, and commercialization.

Sliced just for you:

  • 🤖 Tencent testing DeepSeek to add China’s hottest AI technology to its Weixin super app

  • 🧠 Are We Outsourcing Our Brains? A Microsoft & Carnegie Mellon study on AI’s impact on critical thought

  • 🏥 University of Hong Kong develops AI tool for fast, precise cancer detection

  • 🇰🇷 South Korea aims to secure 10,000 GPUs for national AI computing center

  • 🚫 South Korea bans downloads of China’s DeepSeek AI over data security concerns

Meet Tanka, the AI Messenger with Memory for Teams. Unlike traditional messaging apps that are forgetful, Tanka remembers, learns, and evolves.

❤️ What You’ll Get in This Exclusive Webinar
✅ How AI memory transforms team collaboration and decision-making.
✅ Live Demo: Watch Tanka in action.
✅ Why AI memory is the next big step for AGI—and how Tanka does it differently.
✅ Real-world impact: See how teams are already saving hours with Tanka.
✅ Q&A + Exclusive Early Access & Offers for attendees!

Also, Tanka is actively hiring! Register for the webinar to learn more the various roles available in AI research, engineering, PM, and GTM (Silicon Valley & Singapore).

Tencent is testing the integration of DeepSeek’s AI technology into its Weixin super app, giving its 1.3 billion active users access to advanced AI-powered search features. The beta version allows users to click an “AI Search” option to use DeepSeek alongside Tencent’s own Hunyuan model. DeepSeek-R1, an open-source reasoning model, has gained attention for its capabilities comparable to those of OpenAI, Anthropic, and Google, but with significantly lower training costs. Tencent has also incorporated DeepSeek-R1 into its Yuanbao AI chatbot. This move aligns with a broader industry trend, as major Chinese tech firms, including Alibaba, Huawei, and Baidu, as well as China’s leading telecom operators, rapidly integrate DeepSeek’s models into their platforms. Smartphone brands such as Huawei, Oppo, Honor, and Vivo are also adopting DeepSeek to enhance AI features in their devices, underscoring the growing influence of domestic AI technology in China’s digital ecosystem.

A joint study by Microsoft Research and Carnegie Mellon University explores how generative AI impacts critical thinking among knowledge workers. Surveying 319 professionals, researchers gathered 936 real-world examples of AI-assisted tasks, finding that while AI streamlines work, it can also erode critical thought. The study highlights that users who over-trust AI are less likely to scrutinize outputs, particularly in routine tasks. Meanwhile, professionals with domain expertise remain more skeptical and engaged in refining AI-generated content. The shift from content creation to AI oversight is evident, as workers now focus on curating and verifying AI-generated material rather than composing from scratch. However, barriers like time constraints, workplace incentives, and lack of subject knowledge hinder deep review. The study calls for AI systems with built-in prompts and verification mechanisms to encourage critical engagement, ensuring AI remains a tool for augmentation rather than replacement of human reasoning.

Researchers at the University of Hong Kong (HKU) have developed Cyto-Morphology Adversarial Distillation (CytoMAD), an AI-powered imaging tool designed to enhance the speed and accuracy of cancer diagnosis. Unlike traditional imaging methods that require time-consuming staining and labeling, CytoMAD employs generative AI to analyze single-cell structures without additional processing, improving image clarity and extracting previously undetectable cellular information. Integrated with a microfluidic system, it enables rapid, cost-effective imaging, supporting clinicians in assessing tumor characteristics and metastasis risks. The tool addresses challenges like batch effect inconsistencies and the reliance on expensive fluorescence markers by providing a label-free alternative that maintains diagnostic precision. Its high-throughput optical imaging technology allows for millions of cell images to be processed daily, accelerating AI model training and medical research applications. Beyond lung cancer diagnostics, CytoMAD holds potential for drug discovery, predictive healthcare, and early disease detection, with plans for a three-year clinical trial to validate its effectiveness in real-world medical settings.

South Korea has announced plans to secure 10,000 high-performance GPUs in 2025 as part of its national AI computing initiative. The government is pursuing public-private cooperation to accelerate the establishment of its AI computing center, positioning itself as a key player in the global AI race. The move follows new U.S. regulations restricting AI chip exports, though South Korea remains exempt from these limitations. While specific GPU models and budget details are still under discussion, they are expected to be finalized by September. Nvidia currently dominates the GPU market, but competitors like AMD and Intel, as well as OpenAI’s move towards in-house chip development, could influence South Korea’s procurement strategy. Meanwhile, emerging Chinese AI firms are focusing on computational efficiency to bridge the performance gap with U.S. processors. This initiative highlights the intensifying global competition for AI infrastructure and computational power.

South Korea has banned new downloads of China’s DeepSeek AI chatbot due to concerns over personal data protection, with authorities stating that access will be restored once the app complies with local regulations. The chatbot saw a surge in popularity, reaching over a million weekly users in the country before being restricted. Several South Korean government agencies had already prohibited employees from downloading DeepSeek on work devices, citing national security risks. Similar measures have been taken by other nations, including Taiwan, Australia, and Italy, while lawmakers in the U.S. are pushing for a federal ban. DeepSeek’s rapid rise and its competitive AI capabilities, reportedly more cost-efficient than U.S. models, have sparked global debate over AI governance, data security, and technological competition between China and the West.

🛠️ AI tools updates

The New York Times is expanding its use of AI across both product and editorial teams, introducing internal tools aimed at improving workflow efficiency. The company has developed a proprietary AI tool called Echo for summarization and newsroom support while also granting access to AI-driven programs like GitHub Copilot, Google’s Vertex AI, and OpenAI’s API. AI will assist with tasks such as SEO headlines, social media copy, summaries, and brainstorming but will not be used to draft full articles or significantly alter content. Editorial guidelines emphasize caution around copyright risks, source protection, and ethical considerations, prohibiting the use of AI to bypass paywalls or publish AI-generated images and videos. Despite enthusiasm from leadership, some staff remain skeptical about AI’s potential impact on creativity and accuracy. The move comes as the Times remains embroiled in a lawsuit against OpenAI, accusing it of unauthorized use of Times content for model training, highlighting the tension between traditional journalism and emerging AI technologies.

💵 Venture Capital updates

Safe Superintelligence Inc. (SSI), the AI startup co-founded by former OpenAI chief scientist Ilya Sutskever, is reportedly raising over $1 billion at a $30 billion valuation. Greenoaks Capital Partners is leading the round, marking a significant jump from the company’s previous $1 billion funding on a $5 billion valuation last year. Founded in June 2024, shortly after Sutskever’s departure from OpenAI, SSI aims to develop AI models with a strong emphasis on safety. The company, co-founded by Daniel Gross and Daniel Levy, has yet to launch a product but is focused solely on building safe superintelligence. Investors are backing the venture largely due to Sutskever’s reputation and the startup’s novel approach, even as the AI investment landscape continues to heat up with massive funding rounds for competitors like OpenAI and Anthropic.

UK-based AI infrastructure startup Ori Industries has secured $140 million to expand its compute-as-a-service business, providing cloud-based access to high-performance computing for AI model training. The funding comes amid a surge in European investment in AI infrastructure, with the EU, France, and the UK committing billions to data center development. Ori is set to roll out access to Nvidia’s cutting-edge H200 and GB200 superchips in the UK, aiming to strengthen Europe’s AI capabilities against competition from the US and China. Additionally, the company has received an undisclosed investment from Saudi Aramco’s VC arm to establish a subsidiary in Saudi Arabia. The startup, which had previously raised $18.1 million, is planning a larger funding round set to close in 2025. This investment aligns with a broader trend of European startups ramping up AI infrastructure, including Nscale, which raised $155 million and pledged £2.5 billion for data center expansion.

🫡 Meme of the day

⭐️ Generative AI image of the day