October 1, 2025

news  
DeepSeek V3.2-Exp Launches, Nvidia, AMD Stocks Rebound

Developments in artificial intelligence continue to accelerate across various sectors. In China, AI startup DeepSeek has released an experimental V3.2-Exp model, which significantly improves training efficiency and reduces costs by over 50 percent. This model incorporates a new 'DeepSeek Sparse Attention' mechanism designed to better process long inputs and is optimized for domestic hardware like Huawei's Ascend chips, supporting China's drive to reduce reliance on foreign technology. Meanwhile, Google is enhancing Google Drive with an AI-powered feature to protect against ransomware, acting as a final defense for common file types by detecting rapid encryption and allowing users to restore previous versions. In the healthcare industry, Philips' Chief Innovation Officer, Shez Partovi, discussed on BBC Asia how AI integrated into medical devices assists doctors by handling tasks like image analysis, enabling them to spend more time with patients. On the financial front, US stock markets saw a rebound in AI-related stocks, with Nvidia and AMD experiencing gains, reflecting ongoing optimism about AI infrastructure spending despite recent volatility. The broader implementation of AI in business is also a focus, with a Boston Consulting Group report highlighting five key principles for successful AI adoption in marketing and sales, emphasizing strategy, data, talent, organizational change, and measurable value. In parallel, US Senators have proposed bipartisan legislation to establish an AI risk evaluation program within the Department of Energy, aiming to track safety concerns related to national security and civil liberties. Peridio has also introduced Avocado OS, a free, open-source system designed to simplify the development of edge AI devices by enabling Linux-based applications to run on various chipsets, including Nvidia's. Finally, the potential of advanced AI, such as agentic AI, is being explored in fields like Environment, Health, and Safety (EHS), though human oversight remains critical for ensuring safety and reliability. Databricks has also launched a new AI solution for cybersecurity to help organizations combat sophisticated threats by unifying security data and enabling AI-driven risk detection and response.

Key Takeaways

  • DeepSeek has launched an experimental AI model, V3.2-Exp, which enhances efficiency and cuts costs by over 50 percent, utilizing a new 'DeepSeek Sparse Attention' mechanism.
  • DeepSeek's V3.2-Exp model is optimized for Chinese hardware, including Huawei's Ascend chips, as part of an effort to reduce reliance on Nvidia's technology.
  • Google Drive now features an AI-powered ransomware protection tool that detects rapid file encryption and allows users to restore previous versions.
  • Philips is integrating AI into medical devices to help doctors with tasks like image analysis, allowing for more patient-facing time.
  • US stock markets saw AI-related stocks, including Nvidia and AMD, rebound, indicating continued investor interest in AI infrastructure.
  • A Boston Consulting Group report identified five key principles for successful AI implementation in marketing and sales, stressing strategy, data, talent, and organizational change.
  • US Senators have proposed a bill to create an AI risk evaluation program within the Department of Energy to monitor national security and civil liberties concerns.
  • Peridio's open-source Avocado OS aims to simplify edge AI development, allowing Linux applications to run on diverse chipsets like Qualcomm and Nvidia.
  • Agentic AI shows promise for EHS management but requires human oversight for safety and reliability.
  • Databricks has introduced a new AI solution for cybersecurity to unify security data and improve threat detection and response.

China's DeepSeek releases experimental V3.2-Exp AI model

China's AI startup DeepSeek has launched an experimental version of its V3 foundation model, called V3.2-Exp. This new model improves training and processing efficiency and cuts costs by over 50 percent compared to older versions. It uses a new 'DeepSeek Sparse Attention' mechanism to better handle long inputs. DeepSeek's V3.1-Terminus model was previously ranked among the top global open-source AI models, tying with OpenAI's gpt-oss-120b.

DeepSeek's new AI model V3.2-Exp boosts efficiency

Chinese AI company DeepSeek released an experimental version of its V3.1-Terminus model, named V3.2-Exp. This new model focuses on increasing efficiency and lowering costs for AI systems. It features 'DeepSeek Sparse Attention' (DSA), which helps the AI process long documents and conversations more effectively and halves the cost of running the AI. Experts believe this makes powerful AI more accessible to developers and smaller companies.

DeepSeek AI model optimized for Chinese chips

Chinese AI firm DeepSeek launched its V3.2-Exp AI model, which is optimized for domestic hardware like Huawei's Ascend chips and CANN software. This move supports China's goal of reducing reliance on Nvidia's CUDA. The model uses a sparse attention mechanism to lower costs for processing long text. Companies like Cambricon and Hygon are also updating their systems to support this new model, showing a coordinated effort to build China's AI ecosystem.

Philips CIO discusses AI's role in healthcare on BBC Asia

Shez Partovi, Chief Innovation Officer at Philips, spoke with BBC Asia about how artificial intelligence is being integrated into medical devices. He explained that AI helps doctors spend more time with patients by handling tasks like analyzing medical images. The interview also touched on the potential impact of AI in healthcare, emphasizing its role in supporting physicians rather than replacing them.

Philips CIO Shez Partovi discusses AI's role in healthcare

Shez Partovi, Chief Innovation Officer at Philips, appeared on BBC Asia to discuss artificial intelligence in healthcare. He highlighted how AI integrated into medical devices allows doctors to dedicate more face-to-face time with patients. The interview explored the growing use of AI in medicine and its potential to enhance patient care.

US stock markets rise as AI stocks rebound

The S&P 500 and Nasdaq saw slight gains on Monday as stocks related to artificial intelligence recovered from a previous week's selloff. Nvidia, Advanced Micro Devices, and Micron Technology all experienced increases. Electronic Arts also jumped after announcing a major acquisition deal. Despite recent volatility, analysts remain optimistic about AI infrastructure spending, though they note potential risks like market concentration.

Five principles for successful AI in marketing and sales

A Boston Consulting Group (BCG) survey found that only 25% of companies successfully use AI in marketing, sales, and pricing. The report identifies five key principles for success: aligning AI with business strategy, building a strong data foundation, developing talent, managing organizational change, and focusing on measurable business value. Companies that struggle often lack one or more of these elements, indicating that effective AI implementation requires more than just technology.

Google Drive adds AI protection against ransomware

Google has introduced a new AI-powered feature for Google Drive to protect against ransomware attacks. This tool acts as a final defense, especially for common file types like Microsoft Office documents and PDFs. When it detects the rapid encryption behavior typical of ransomware, it suspends syncing to prevent corrupted files from overwriting good ones. Users can then easily restore their files to a previous state.

US Senators propose AI safety evaluation bill

Senators Josh Hawley and Richard Blumenthal introduced bipartisan legislation to create an AI risk evaluation program within the Department of Energy. This program aims to track AI safety concerns related to national security, civil liberties, and labor. Developers of advanced AI systems would need to provide product information to the DOE before deployment. The bill requires annual reports to Congress on federal oversight recommendations for AI systems.

Peridio's Avocado OS simplifies edge AI migration

Peridio demonstrated its free, open-source Avocado OS at Microelectronics UK 2025, showcasing its ability to simplify development for edge AI devices. The OS allows Linux-based applications to run seamlessly on different hardware, including Qualcomm and Nvidia chipsets. Avocado OS offers a secure, scalable, and composable architecture, resolving the dilemma between using difficult-to-scale distributions and complex custom builds.

Agentic AI in EHS requires human oversight

Agentic AI, an advanced form of artificial intelligence, is poised to transform Environment, Health, and Safety (EHS) management. While these AI agents can plan and act autonomously, they require human validation to ensure safety and reliability. A 'human-in-the-loop' approach is crucial in EHS to prevent errors that could lead to serious harm. The partnership between agentic AI and human judgment will be key to future success in EHS.

Databricks launches AI solution for cybersecurity

Databricks has released Data Intelligence for Cybersecurity, a new solution to help organizations combat modern and AI-driven cyber threats. It unifies security data and uses AI for earlier risk detection and faster response. The solution includes 'Agent Bricks' for building AI apps that can take governed actions. This aims to address challenges caused by increasingly sophisticated attacks and fragmented data, providing real-time intelligence and flexibility.

Sources

AI model DeepSeek V3.2-Exp AI efficiency AI costs Sparse Attention long inputs open-source AI OpenAI AI hardware Huawei Ascend CANN software Nvidia CUDA Cambricon Hygon AI ecosystem AI in healthcare Philips medical devices medical imaging patient care physician support US stock markets AI stocks Nvidia Advanced Micro Devices Micron Technology AI infrastructure AI in marketing AI in sales AI in pricing business strategy data foundation AI talent organizational change business value Google Drive AI protection ransomware AI security AI safety AI risk evaluation Department of Energy national security civil liberties labor federal oversight edge AI Peridio Avocado OS edge AI devices Linux Qualcomm Nvidia chipsets Agentic AI EHS management human oversight human-in-the-loop AI in cybersecurity Databricks cyber threats AI-driven threats risk detection AI apps governed actions real-time intelligence